Post on 08-Sep-2020
Projekt współfinansowany ze środków Unii Europejskiej w ramach Europejskiego Funduszu Społecznego
ROZWÓJ POTENCJAŁU I OFERTY DYDAKTYCZNEJ POLITECHNIKI WROCŁAWSKIEJ
Wrocław University of Technology
Control in Electrical Power Engineering
Przemysław Janik, Tomasz Sikorski
POWER QUALITY
ASSESSMENT
Wrocław 2011
Wrocław University of Technology
Control in Electrical Power Engineering
Przemysław Janik, Tomasz Sikorski
POWER QUALITY ASSESSMENT
Compressor Refrigeration Systems, Heat Pumps,
Wrocław 2011
Copyright © by Wrocław University of Technology
Wrocław 2011
Reviewer: Zbigniew Leonowicz
ISBN 978-83-62098-64-4
Published by PRINTPAP Łódź, www.printpap.pl
Content
1. INTRODUCTION ....................................................................................................................................... 4
2. OVERVIEW OF DISTURBING PHENOMENA AND INDICES ......................................................................... 6
2.1. MOTIVATION FOR SYSTEMATICAL APPROACH .................................................................................................. 6
2.2. VOLTAGE SAG ........................................................................................................................................... 7
2.3. BRIEF INTERRUPTIONS ................................................................................................................................ 8
2.4. VOLTAGE SWELLS ...................................................................................................................................... 9
2.5. TRANSIENTS ........................................................................................................................................... 10
2.6. UNBALANCE ........................................................................................................................................... 10
2.7. HARMONIC DISTORTION ........................................................................................................................... 11
2.8. FLICKER ................................................................................................................................................. 13
2.9. POWER QUALITY STANDARDS .................................................................................................................... 14
2.10. LITERATURE ........................................................................................................................................... 15
3. RESONANCES ........................................................................................................................................ 16
3.1. PROBLEM FORMULATION .......................................................................................................................... 16
3.2. ASSESSMENT OF SERIES RESONANCE ............................................................................................................ 18
3.3. ASSESSMENT OF PARALLEL RESONANCE........................................................................................................ 21
3.4. LITERATURE ........................................................................................................................................... 26
4. ANALOG FILTER DESIGN METHODOLOGY FOR HARMONIC CANCELATION ............................................ 26
4.1. PROBLEM FORMULATION .......................................................................................................................... 26
4.2. HIGHER ORDER HARMONICS ORIGINATING IN NONLINEAR LOADS ...................................................................... 28
4.3. CURRENTS DRAWN BY A NON-SINUSOIDAL LOAD, GENERAL CHARACTERISTIC ....................................................... 29
4.4. BASIC DEPENDENCIES IN THE RESONANT CIRCUIT SELECTED FOR THE FILTER ......................................................... 32
4.5. PROPOSED DESIGN FOR EFFICIENT HARMONIC FILTRATION ............................................................................... 35
4.6. COMPUTATION OF FILTRATION PARAMETERS ................................................................................................ 36
4.7. SIMULATION OF NONLINEAR LOAD WORKING WITH A FILTER ............................................................................ 39
4.8. LITERATURE ........................................................................................................................................... 42
5. FOURIER TECHNIQUES FOR SPECTRAL ANALYSIS ................................................................................... 42
5.1. DISCRETE TIME FOURIER TRANSFORM DEFINITION ........................................................................................ 43
5.2. DISCRETE FOURIER TRANSFORM DEFINITION ................................................................................................ 45
5.3. DISCRETE FOURIER TRANSFORM EXPRESSED USING TRIGONOMETRIC FUNCTIONS ................................................. 47
5.4. PROPERTIES OF DFT ................................................................................................................................ 56
5.5. SHIFTING THEOREM IN DFT ....................................................................................................................... 58
3
5.6. INVERSE DISCRETE FOURIER TRANSFORM ..................................................................................................... 62
5.7. DFT LEAKAGE (AMBIGUITY) AND ITS MINIMIZATION ...................................................................................... 62
5.8. WINDOWS TYPES AND THEIR PROPERTIES ..................................................................................................... 70
5.9. RESOLUTION OF THE DFT, FILLING WITH ZEROS, SAMPLING IN THE FREQUENCY DOMAIN ....................................... 72
5.10. LITERATURE ........................................................................................................................................... 74
6. SYMMETRICAL COMPONENTS FOR CURRENT AND VOLTAGE UNBALANCE ASSESSMENT ...................... 75
6.1. MATHEMATICAL BACKGROUND ................................................................................................................. 76
6.1.1. Complex transformation parameter a........................................................................................... 76
6.1.2. Symmetrical components definition .............................................................................................. 77
6.1.3. Symmetrical components of voltage, current, impedance and admittance.................................. 79
6.1.4. Basic properties of symmetrical components ................................................................................ 80
6.2. OHM’S LOW FOR SYMMETRICAL COMPONENTS ............................................................................................. 81
6.3. KIRCHHOFF’S CURRENT LOW AND VOLTAGE LOW FOR SYMMETRICAL COMPONENTS .............................................. 83
6.4. FILTERING OF SYMMETRICAL COMPONENTS .................................................................................................. 84
6.4.1. Zero sequence filtering .................................................................................................................. 85
6.4.2. Positive and negative sequence components filtering .................................................................. 86
6.5. DETERMINATION OF UNBALANCE WITH ACCORDANCE TO INTERNATIONAL STANDARDS .......................................... 90
6.5.1. Unbalance factor ........................................................................................................................... 90
6.5.2. Evaluation according to standards ................................................................................................ 91
6.6. LITERATURE ........................................................................................................................................... 92
7. ADVANCED TOPIC. DETERMINATION OF TRANSIENT PARAMETERS IN WIND ENERGY CONVERSION
SYSTEM .......................................................................................................................................................... 92
7.1. MOTIVATION AND PROBLEM FORMULATION ................................................................................................. 92
7.2. PRONY METOD ....................................................................................................................................... 93
7.3. NONLINEAR REGRESSION METHOD .............................................................................................................. 96
7.4. WAVELET TRANSFORM............................................................................................................................. 97
7.5. SIMULATION OF INDUCTION GENERATOR WITH CAPACITORS ............................................................................. 99
7.6. PARAMETERS OF TRANSIENTS COMPUTED USING PRONY ALGORITHM AND NONLINEAR REGRESSION ...................... 100
7.7. CONCLUDING REMARKS .......................................................................................................................... 105
7.8. LITERATURE ......................................................................................................................................... 105
8. COMPUTATIONAL APPLICATION OF POWER QUALITY ASSESSMENT: DIPS .......................................... 107
8.1. METHODS OF VOLTAGE MAGNITUDE ESTIMATION ........................................................................................ 109
8.2. APPLICATION OF RMS TREND CALCULATION IN MATLAB ................................................................................ 113
8.3. COMPARISON OF THE METHODS ............................................................................................................... 119
8.4. DIP DURATION ...................................................................................................................................... 121
8.5. PHASE-ANGLE JUMP .............................................................................................................................. 125
4
8.6. MONITORING MODE IN VOLTAGE DIPS ASSESSMENT ..................................................................................... 128
8.7. DIP TRANSFER IN POWER SYSTEMS ............................................................................................................ 132
8.8. LITERATURE ......................................................................................................................................... 137
9. COMPUTATIONAL APPLICATION OF POWER QUALITY ASSESSMENT: HARMONICS ............................. 139
9.1. HARMONICS DISTORTION AND FOURIER SERIES PARAMETERIZATION ................................................................ 140
9.2. INVESTIGATED PHENOMENA .................................................................................................................... 143
9.3. “LINEAR” SPECTRUM ESTIMATION USING FFT: ............................................................................................ 145
9.4. HARMONICS AND SUBDIVISION INTO EVEN AND ODD HARMONICS ................................................................... 148
9.5. GROUPS OF HARMONICS, INTERHARMONIC AND SUHARMONIC ....................................................................... 150
9.6. LITERATURE ......................................................................................................................................... 155
10. ELEMENTS OF POWER QUALITY REPORT ............................................................................................. 156
10.1. POWER FREQUENCY ASSESSMENT ............................................................................................................. 158
10.2. SUPPLY VOLTAGE VARIATION ................................................................................................................... 159
10.3. FLICKER SEVERITY .................................................................................................................................. 160
10.4. VOLTAGE UNBALANCE ASSESSMENT .......................................................................................................... 161
10.5. HARMONICS ASSESSMENT ....................................................................................................................... 162
10.6. SUMMARY OF THE REPORT ...................................................................................................................... 165
10.7. LITERATURE ......................................................................................................................................... 165
5
1. Introduction
Power Quality (PQ) topics are of outmost importance in our world strictly dependent on
continuous supply of electrical energy.
The field of Power Quality has no strictly defined borders as continuously more and more
topics are being discussed in this area. A brief glance at headings of the international
conferences justifies the presumption. The International Conference on Harmonics and
Quality of Power (ICHQP) and The International Conference on Renewable Energies and
Power Quality ICREPQ are two good examples reflecting the research activities across
continents.
From the technical point of view we have a steady growing number of nonlinear loads,
electronic controlled devices saving lamps etc., drawing highly non sinusoidal currents. On
the other hand, computers, electronic control systems and protection devices are vulnerable
and prone to power quality disturbances. The electrical network may be seen as an
interconnection between the sinks and the sources of electrical disturbances. Standards and
legal regulations help to find the “golden section” between allowed vulnerability of devices
and allowed distortion levels in the electrical network.
The title of the book indicates strictly its purpose. In the vast area of PQ, it should give a
student an inside look into the theoretical background of Power Quality assessment. So that
reading and understanding of changing standards, procedures and technologies will be
smooth and based on relatively solid foundations. It should help the students to focus on PQ
problems in every day life of electrical engineer working in the deregulated environment.
After the Introduction, a brief overview of basic disturbing phenomena, their origin, effects
and mitigation methods is given in Chapter 2.
In Chapter 3, the reader is given information about a common but dangerous phenomenon
of resonances. The conditions for parallel and series resonance are given along with features
and possible thread for the electrical components.
6
Chapter 4 presents an effective approach to an analog filter design helping to reduce
harmonics. As the filter uses the idea of resonance, it is a natural consequence of the
previous chapter. The theoretical presentation is rounded up with practical implementation
and simulated filtration results.
Chapter 5 is a fulfillment or a logical consequence of the previous one. If we want to cancel
or filter harmonics we should firstly assess their presence in the electrical grid. Fourier
transform is a popular and powerful method in spectral analysis. However, appropriate
computation of spectral components requires from the engineer at least basic knowledge of
DFT properties and features.
Chapter 6 gives us the look into a three phase system. Previously presented topics can be
usefully used for every phase of the three phase system separately, or for all of them
together. Symmetrical components and symmetry as a global look at a three phase system
are inevitable in the assessment of power quality for the results of non symmetry are
disastrous.
Chapter 7 is an example of an advanced approach to signal parameters estimation in a
system with a wind generator. The complete process has been shown. Starting with the
introduction and motivation through methodology and algorithms up to the discussion of
results.
Chapter 8 is dedicated to undervoltage event member, called voltage dips (or sags). A proper
classification of this event requires an estimation of voltage magnitude as well as time
duration during the dip. This chapter introduces mathematical backgrounds as well as a
practical application of algorithms in Matlab environment with comparison and examples. In
order to emphasize rules of the dips transfer in the power systems an example of a power
system model is presented.
Chapter 9 is dedicated to application of Fast Fourier Transform for spectrum estimation.
Selected issues of harmonics, interharmonics and harmonics group are introduced.
Demonstration algorithms in Matlab and its results are presented on the basis of real
measured signal of dc-arc furnace plant.
The aim of chapter 10 is to present an example of power quality report including the
assessment of mentioned crucial parameters. Measurements were done in the main point of
the supply of the factory which has an electronic assembly line including robots and power
electronics.
7
2. Overview of disturbing phenomena and indices
The overview of disturbing phenomena starts with a brief, systematized motivation
formulation of the interest in power quality, generally. This text is mainly based on [2].
In the context of power quality, a disturbance is a temporary deviation from the steady state
of waveform caused by faults of brief duration or by sudden changes in the power system.
The disturbances considered by the International Electromagnetic Commission include
voltage dips, brief interruptions, voltage increases, impulsive transients and oscillatory
transients [2].
Nevertheless, waveform distortions (e.g. harmonics), unbalance, voltage fluctuation and
flicker indicate clearly the deterioration of power quality. They are also mentioned in
standards, and therefore will be introduced.
2.1. Motivation for systematical approach
Before a detailed presentation of disturbances types is given, there is a need for justification
of the systematic approach to Power Quality. The individual topics discussed in this field are
rather not new. What is new is the system approach in the engineering, seeking
interconnections and dependences between various, seemingly independent phenomena. A
global assessment is needed to operate the electrical system safely and reliably. That is why
disturbances have been grouped into classes which are simultaneously controlled and
checked at various points of the electrical system.
Generally, there are four major reasons for the systematic approach and growing interest in
Power Quality [1].
• Load equipment is more sensitive to power quality variations than equipment
applied in the past. Many new load devices contain microprocessor based controls
and power electronic devices that are sensitive to many types of disturbances.
• The increasing emphasis on overall power system efficiency has resulted in a
continued growth in the application of devices such as high-efficiency, adjustable
speed motor drives and shunt capacitors for power factor correction to reduce
losses. This results in increased harmonic levels on power systems and has many
people concerned about the future impact on system capabilities.
8
• Increased awareness of power quality issues by the end users. Utility customers are
becoming better informed about such issues as interruptions, sags, and switching
transients and challenge the utilities to improve the quality of power delivered.
• Many things are now interconnected in a network. Integrated processes mean that
the failure of any component has much more important consequences.
Interestingly, the equipment installed to increase the productivity is also often the
equipment that suffers the most from common power disruptions. And the equipment is
sometimes the source of additional power quality problems.
2.2. Voltage sag
Voltage sag (sometimes called a dip) is defined as a sudden reduction of the voltage, ranging
between 10% and 90% of the nominal voltage. The duration of a sag is from 0.5 cycle, up to
several seconds. An example of a symmetrical voltage sag in a three phase system is shown
in Figure 2.1.
Common sources of voltage sags are switching operations in the electrical network, changing
transformer taps, changing configuration of the grid, switching heavy loads, especially
electrical drives. Faults often affect a wide area causing sags lasting as long as the clearing of
the fault. the simplified principle of the phenomenon is shown in Figure 2.2. The severity of a
particular sag originating in a fault may be given by a simple formula
= ⋅+Z
sag
S Z
ZU E
Z Z (2.1)
The effect of voltage dips on equipment depends on both parameters of a sag, its duration
and magnitude. Computer manufacturers developed a standard correlating the duration and
severity of a sag with electronic equipment immunity, called CBEMA. The possible effect on
equipment are extinction of discharge lamps, incorrect operation of control devices, speed
variations and stopping of motors, tripping of contactors, computer miss operation [2]. And
many more.
There is no simple and cheap protection against voltage sags. In case of small loads
uninterrupted power supplies UPS are quite efficient. The use of power conditioners or even
rebuilding of the electrical grid to reduce voltage losses on impedances are suggested for
greater loads.
9
0 0.02 0.04 0.06 0.08 0.1 0.12 0.14 0.16 0.18-1
-0.5
0
0.5
1
time [s]
amplitu
de [V
]
Figure 2.1 Three phase voltage sag, symmetrical case
Figure 2.2 Simplified model of a voltage divider representing the proliferation of sags
2.3. Brief Interruptions
No voltage at all may be considered as the most severe sag at all. The effects on equipment
is an immediate shut-down and an unexpected stop of operation. Lost data and interrupted
industrial processes may lead to huge financial losses.
An example of a three phase voltage interruption is shown in Figure 2.3.
The prime protection procedure against short interruptions is installation of uninterrupted
power supplies.
10
0 0.02 0.04 0.06 0.08 0.1 0.12 0.14 0.16 0.18-1
-0.8
-0.6
-0.4
-0.2
0
0.2
0.4
0.6
0.8
1
time [s]
amplitu
de [pu]
Figure 2.3 Voltage interruption in a three phase system.
2.4. Voltage swells
Voltage swell is the opposite of a sag. A swell is defined as an increase to between 1.1 and
1.8 in pu for durations from 0.5 cycle to several seconds. Swells are also characterized by its
magnitude and duration. A three phase case is shown in Figure 2.4. An unsymmetrical
voltage swell is often present in un-faulted phases when a single phase fault occurs [2].
0.02 0.04 0.06 0.08 0.1 0.12 0.14
-1
-0.5
0
0.5
1
1.5
time [s]
amplitud
e [p
u]
Figure 2.4 Symmetrical voltage swell in a three phase system
Swells can upset electric controls and electric motor dives, which can trip because of their
build in protection devices.
A possible solution to the problem is the application of uninterruptible power supplies and
conditioners.
11
2.5. Transients
Voltage disturbances shorter than sags or swells are classified as transients and are caused
by sudden changes in the power system. An example of a voltage sag is shown in Figure 2.5.
0.03 0.04 0.05 0.06 0.07 0.08 0.09 0.1
-1
-0.5
0
0.5
1
1.5
time [s]
amplitu
de [p
u]
Figure 2.5 Oscillatory transient in a three phase system
According to their duration, transient overvoltages can be divided into switching surge
(duration in the range of milliseconds), and impulse spike (duration in the range of
microseconds).
Surges are high energy pulses arising from power system switching disturbances, either
directly or as a result of resonating circuits associated with switching devices. They also
occur during step load changes.
Capacitor switching can cause resonant oscillations leading to an overvoltage some three to
four times of the nominal rating, causing tripping or even damaging protective devices and
equipment. Electronically based controls for industrial motors are particular susceptible to
these transients.
Protection against surges and impulses is normally achieved by surge-diverters and arc-gaps
at high voltages and avalanche diodes at low voltages.
Faster transients in nanoseconds due to electrostatic discharges, an important category of
EMC, are not normally discussed under power quality.
2.6. Unbalance
Unbalance describes a situation in which either the voltages of a three-phase voltage source
are not identical in magnitude or the phase differences between them are not 120 electrical
12
degrees between them, or both. An example of unbalanced three phase voltages is shown in
.
0 0.02 0.04 0.06 0.08 0.1 0.12 0.14 0.16-1
-0.5
0
0.5
1
time [s]
amplitu
de [p
u]
Figure 2.6 Voltage unbalance
The degree of unbalance is usually defined by the proportion of negative and zero sequence
components. This subject is discussed extensively in the chapter dedicated to unbalance.
The main causes of unbalance are single phase loads and un-transposed overhead
transmission lines.
An electrical machine operating on an unbalanced supply will draw a current with a degree
of unbalance several times that of the supply voltage. As a result, the three phase currents
may differ considerably and a temperature rise will take place in the machine.
Motors and generators may be fitted with protection to detect extreme unbalance. If the
supply unbalance is sufficient, the single phase protection may respond to the unbalanced
currents and trip the machine.
2.7. Harmonic distortion
Waveform distortion is generally discussed in terms of harmonics. This is a fast subject for
which special books have been dedicated [3]. At this point only a general introduction will be
given. Three phase signal distorted by harmonics is shown in Figure 2.7.
13
0 0.005 0.01 0.015 0.02 0.025 0.03 0.035 0.04-1
-0.5
0
0.5
1
1.5
time [s]
amplitu
de [p
u]
Figure 2.7 Three phase signal distorted by harmonics
Generally, any non-linear load is a source of current harmonics in the electrical system. The
harmonic sources can be grouped into three categories according to their origin, size and
predictability:
- small and predictable (domestic and residential appliances)
- large and random (arc furnaces, welders)
- large and predictable (static converters, HVDC transmission)
The main detrimental effects of harmonics are:
- maloperation of control devices, signaling systems, protective relays
- extra losses in capacitors, transformers and rotating machines
- additional noise from motors and other apparatus
- telephone interference
- the presence of power factor correction capacitors and cable capacitance is a
potential reason for shunt and series resonances resulting in voltage
amplification
To keep the harmonic voltage content within the recommended levels the main solutions
are:
- the use of high pulse rectification (e.g. in HVDC transmission systems)
- passive filters, either tuned to individual frequencies or of the band pass type.
- active filters and conditioners
14
2.8. Flicker
Flicker has been described in [4] as:
“the impression of fluctuating luminance or color occurring when the frequency of the
variation of the light stimulus lies between a few hertz and the fusion frequency of images.
This is a very loose definition considering that “the fusion frequency of images” varies from
person to person and depends on many factors.
Fluctuation of the system voltage (more specifically in the r.m.s. value) can cause perceptible
(low frequency) light flicker depending on the magnitude and frequency of the variation.
Power system engineers call this type of disturbance “voltage flicker” but often it is just
shortened to “flicker”.
A simple flicker in a three phase system is shown in Figure 2.8. It represents a simple case of
a voltage flicker, when the a.c. voltage is modulated (amplitude modulation) by a sine wave
seen as the envelope of the voltage waveform.
0 0.02 0.04 0.06 0.08 0.1 0.12 0.14 0.16-1.5
-1
-0.5
0
0.5
1
1.5
time [s]
aplitud
e [pu]
Figure 2.8 Voltage flicker in a three phase system
The general mathematical expression for a flicker is given by
( ) ( ) ( )0 0 0 0
1
sin sin=
= + + +
∑M
fi fi fi
i
v t t tα α β ω ϕ ω ϕ (2.2)
and can be interpreted as modulation by a sine wave with relatively low frequency.
Non-periodic events can also cause a perceptible light flicker. Any potentially perceptible
change in brightness should therefore be termed light flicker, broadening the given
definition further by extending the lower frequency limit to non-periodic disturbances.
The main causes of flicker are loads drawing large and highly variable currents. Due to the
impedance of the power system (generators, transformers and transmission lines) these
15
changes produce amplitude modulation of the voltage at the load bus and even at remote
busses. Another common source of flicker is the starting of electric motors. Their operation
in application that require an irregular torque is problematic. Motor applications range from
household appliances up to more powerful appliances such as heat pumps or rolling mills.
These flicker sources lead to electric lamp luminance fluctuations by means of amplitude
modulation of the supply voltage.
The flickering of electric light causes annoyance to human observers. It reduces the live of
electronic devices, incandescent and fluorescent devices, malfunction of phase-locked-loops
and loss of synchronism in uninterruptible power supplies, maloperation of electronic
controllers and protection devices.
2.9. Power Quality Standards
Te development of standards and guidelines is centered around the following objectives [5]:
- description and characterization of the phenomena
- major sources of power quality problems
- impact on other equipment and on the power system
- mathematical description of the phenomena using indices or statistical
analysis to provide a quantitative assessment of its significance
- measurement techniques and guidelines
- emission limits for different types and classes of equipment
- immunity or tolerance level of different types of equipment
- testing methods and procedures for compliance with limits
- mitigation guidelines
The internationally recognized organization for standardization is the International
Electrotechnical Commission which is based in Geneva. IEC has defined a series of standards,
called Electrotechnical Capability, to deal with power quality issues. This series is published
in separate parts according to the following structures.
1. General (IEC 61000-1-x) the general section introduces and provides fundamental
principles on EMC issues and describes the various definitions and terminologies
used in the standards
16
2. Environment (IEC 61000-2-x) this part describes and classifies the characteristics of
the environment or surroundings where equipotent will be used. It also provides
guidelines on compatibility levels for various disturbances.
3. Limits (IEC 61000-3-x) this section defines the maximum levels of disturbances
caused by equipment or appliances that can be tolerated within the power system. It
also defines the immunity limits or equipment sensitive to EMC disturbances.
4. Testing and Measurement Techniques (IEC 61000-4-x)these provide guidelines on the
design of equipment for measuring and monitoring of power quality disturbances.
They also outline the equipment.
5. Installations and Mitigation Guidelines (IEC 61000-5-x) this section provides
guidelines on the installation techniques to minimize emission and as well as to
strengthening immunity against EMC disturbances. It also describes the use of
various devices for solving power quality problems.
6. Generic Standards (IEC 61000-6-x) these include the standards specific to certain
category of equipment or for certain environments. They contain both emission
limits and immunity levels standards
Some of the other organizations who have developed their own standards are CENELEC, UIE,
IEEE, ANSI, NEMA. These standards are usually very much application based on or specific to
a certain environment.
The standard EN 50160 Voltage Characteristics in Public Distribution Systems is widely used
in Europe and directly build-in into commercial power quality metering devices. This
standard, or its pats is also included in legal regulations concerned with power quality.
2.10. Literature
[1] R. C. Dugan, M. F. McGranaghan, H. W. Beaty, Electric Power Systems Quality,
McGraw-Hill, New York, 1996
[2] J. Arrilaga, N. R. Watson, Power System Quality Assessment, John Wiley and Sons,
Chichester, 2000
[3] J. Arrillaga, N. R. Watson, Power System Harmonics, Wiley and Sons, Chichester, 2003
[4] IEEE, Standard Dictionary of Electrical and Electronic Terms, Std.100, IEEE, 1984
17
[5] J. Arrillaga, D.A. Bradley, P.S. Bodger, Power System Harmonics, John Wiley & Sons,
1985
3. Resonances
Resonances are a common phenomenon in an electrical network. The occurrence of
resonance is directly connected with other power quality issues. If there are unexpected
harmonics, then the damage to system components may be high due to resonances. On the
other hand, the resonance phenomenon is exploited in filtering devices, to short-circuit the
unwanted harmonics. That’s why the resonance issue coms first. The presentation and the
problem formulation is directly based on [3].
3.1. problem formulation
When the voltage applied to an electrical network containing resistance, inductance and
capacitance is in phase with the resulting current, the circuit is said to be resonant. At
resonance, the equivalent network impedance is purely resistive, since the supplied voltage
and current are in phase. The power factor of a resonant network is unity [1].
Apart from power quality issues, the phenomenon of the resonance is of a great value in
communication engineering. It enables small portions of the communication frequency
spectrum to be selected for amplification independently of the remainder.
However, there is also a potential thread inherent in this phenomenon. If a voltage
resonance or a current resonance is present in an electrical network, it may be dangerous
for system components (damaged insulation, flash over, etc.).
Excess current or voltage may be dangerous for insulation of wires and components.
Especially capacitors and transformers insulation is vulnerable to high voltages. Measuring
transformers may also be damaged. Currents higher than allowed are responsible for
additional mechanical stress put on all current leading elements. There is also additional
heat accelerating the aging process of insulation. There is also a risk of saturation in
transformers. Electrical apparatus, measuring transformers other devices malfunction can
often be led back to unwanted resonances.
Special care is needed during the design process to avoid resonances for normal
operation condition at 50 Hz. The situation becomes more complicated if harmonics and
18
interharmonics are present in current or voltage due to changes of capacitance and
inductance along with frequency. That means, if there is a problem with harmonics, the risk
of resonances is significantly higher. The common source of harmonics are nonlinear devices
of any type, mostly power electronic. Special harmonic filters are designed for them. A
slightly more difficult situation results from transient overvoltages and overcurrents.
Lightning and capacitor switching are very typical examples. Capacitors are used to provide
reactive power to correct the power factor, they are cheap and sufficient, but switching
operations are quite frequent.
A wind generator equipped with compensating capacitors, as an example of a system
component with variable consumption of reactive power is shown in Figure 3.1 [2]. Due to
the variable consumption of reactive power, let us stress it again, the switching operations
are frequent. The parameters of oscillatory transient vary with changing wind conditions and
system parameters. An example of current time curve is shown in Figure 3.2.
IM
TrafoGenerator
Bus Sys Bus 1 Bus 2 Bus 3
Line
Figure 3.1 Induction generator with compensating capacitors
Time [s]0 0.02 0.04 0.06 0.08 0.1
Current [A]
-800
-400
0
400
Figure 3.2 Current resulting from capacitor bank switching in wind generator
There is no doubt about the necessity to avoid resonances in electrical networks. The
systematic presentation of the phenomenon starts with the computation of conditions for
series or parallel resonance. Then, follows the computation of the resonant frequency,
determining the maximal voltage or current values.
Basic ideas for series resonance and parallel resonance will be shown below. All
voltages and currents are complex, unless otherwise noticed.
19
3.2. Assessment of series resonance
Figure 3.3 shows a circuit comprising a coil of inductance L, resistance R and capacitance
connected in series [3].
Figure 3.3 R, L, C elements connected in series
The voltage across the series RLC circuit is given by
R L CU U U U= + + (3.1)
The circuit is at resonance, if 0L CU U+ = , also
00
10L I
Cω
ω
− =
(3.2)
Thus,
1
LCω = (3.3)
and the frequency at resonance
1
Hz2
fLCπ
= (3.4)
Figure 3.4 shows how the reactance and inductance and their algebraically added values
vary with changing frequency.
Figure 3.4 Variations of R, XL, XR with the frequency
20
Since (3.2) is fulfilled at resonance, then the impedance constantZ R= = . This is the
minimum possible value of impedance and the current reaches its maximum.
At frequencies less than ( )0 0 f ω , L CX X< the circuit is capacitive, at frequencies higher
than ( )0 0 f ω , L CX X> the circuit is inductive.
The inductive reactance or capacitive reactance at resonance frequency is called
characteristic impedance (wave impedance) and usually denoted with ρ .
00
1 LL
C Cρ ω
ω= = = (3.5)
Quality factor or Q-factor [1] in series resonance is the ratio of characteristic impedance to
the resistance
QR
ρ= (3.6)
Therefore the Q-factor of capacitor and inductive coil at resonance is equal to
0
0
1L C
LQ Q
R R C
ωω
= = = (3.7)
In other words, the Q-factor is a ratio of voltage across reactive element to the voltage
across resistance. (3.8)
L C
R R
U UQ
U U= = (3.9)
Equation (3.9) indicates other name of Q - circuit magnification factor.
The Q-factor is defined in general using the energetic approach [4]. Let us multiply the
numerator and denominator of (3.9) by 21
2mI
2 2
max0
2 2
1 1
2 22 21 1 ( )
2 2
m mL
Rm m
I L I LW
QW T
I R RT I
ω π π= = = (3.10)
where maxLW is the maximal energy storied in the magnetic field of a coil and ( )RW T energy
lost at resistance during one period T. The Q-factor is multiplied by 2π ratio of maximal
energy storied in an inductive coil to the energy lost in resistance over one period.
Let us assume that the voltage source frequency feeding RLC branch is variable. If the
frequency is 0ω , we say that the circuit is tuned to resonance. If the angular frequency ω is
21
different from 0ω , we say that the circuit is not tuned to resonance that there is a deviation
from the resonant frequency.
In the impedances triangle for a RLC branch, the value of ( )tg ϕ is equal to the ratio
/X L , which characterizes deviation from resonant frequency ξ .
( ) ( ) also X
tg arctgR
ξ ϕ ϕ ξ= = = (3.11)
Therefore, impedance of RLC branch may be expressed as
21j jZe R eϕ ϕξ= + (3.12)
At resonance 0 and Z=Rξ = , also
21 jZe
R
ϕξ= + (3.13)
Equation (3.11) is presented graphically for three chosen values of resistance: 5,10 and 20
Ohms (Figure 3.5).
Figure 3.5 Phase angle ϕ as a function of the deviation from resonant frequency.
Utilising (3.7) we can express the inductance and capacitance as a function of characteristic
impedance
0 0
1 L C
ρω ω ρ
= = (3.14)
and substitute the values to the expression for RLC branch reactance we get
0
0
1X L
C
ωωω ρ
ω ω ω
= − = −
(3.15)
Relative deviation from the resonant frequency δ is given as a ratio of the reactance X and
characteristic impedance ρ .
22
0 0
0 0
f f
f f
ω ωδ
ω ω= − = − (3.16)
Substituting 0/k f f= we can rewrite (3.16) as
1
kk
δ = − (3.17)
Deviation from the resonant frequency, relative deviation from the resonant frequency and
Q-factor are dependent on each other
Qξ δ= (3.18)
For frequencies near the resonant frequency, it is convenient to apply an approximated
expression for δ . The approximated expression is constructed under the assumption that
0 2ω ω ω+ ≈ , then 'δ is given as
( )2 2
0 0 0 0
0 0 0 0
2 2 1 'kω ω ω ω ω ω ω ω
δ δω ω ωω ω ω
− + −= − = = ≈ = − = (3.19)
Figure 3.6 relative deviation from the resonant frequency (true and approximated) as a function of relative
frequency
3.3. Assessment of parallel resonance
Parallel resonance or current resonance may appear in a circuit with parallel GCL branches
(Figure 3.7). For a given complex voltage U .
23
Figure 3.7 Circuit with parallel branches for current resonance
The currents across elements in the branches are given by
1
R
C
L
I GU
I j CU
I jL
ω
ω
=
= = −
(3.20)
The current across capacitor is 90 degrees ahead of the voltage, the current in the coil 90
degrees back after voltage. Both are reactive, only the current across resistance is active.
Parallel resonance is characterized by the fact that at the resonance frequency 0ω currents
across a coil and a capacitor are equal in module but opposite in phase (vector sum is zero).
Figure 3.8 Parallel circuit at resonance, reactive current component equal zero
Current resonance (Figure 3.8) is expressed by
0C LI I+ = (3.21)
also
00
10C U
Lω
ω
− =
(3.22)
Resonant frequency
0
1
LCω = (3.23)
The susceptance of a parallel branch at resonance is zero
24
00
10C LB B B C
Lω
ω= − = − = (3.24)
The voltage across the parallel branch reaches its maximal value in comparison to values
reached not at resonance
00
1
1
IU
GG j C
Lω
ω
= =
+ −
(3.25)
At resonance, the branch is strictly resistive and current is in phase with the voltage.
The characteristic impedance (wave impedance) for a parallel branch is given by
00
1 LL
C Cρ ω
ω= = = (3.26)
The Q-factor for a parallel branch is defined similarly as for a series circuit and given by
2
0 02
1
2mC U
RQ R C
GUω ω
ρ= = = (3.27)
but the expression is different from (3.6). Both equation ((3.6) and (3.27)) are given for a
particular circuit and can not be generalized.
Deviation from resonant frequency in a parallel circuit is given by
B
Gξ = − (3.28)
Admittance as a function of deviation from resonant frequency
21j jY Y e G eϕ ϕξ− −= = + (3.29)
At resonance we have 0 0 00, Y Y Gϕ = = = , therefore the relative admittance and
impedance have the form
2
0
20
1
1
1
Y
Y
Z
Z
ξ
ξ
= +
=+
(3.30)
Assuming changing frequency ω and constant absolute value of current I the Ohm’s law
for absolute values
U Z I= (3.31)
25
Therefore, the characteristic in Figure 3.9 shows the changes of 0/Z Z as a function of the
deviation from resonant frequency which is equal to the curve showing 0/U U .
Figure 3.9 Absolute value of relative impedance as a function of deviation from resonant frequency
Accompanying the characteristic of absolute value of relative impedance the characteristic
of an admittance phase angle is given in Figure 3.10. It has the form of the function arctg
multiplied by minus one.
Figure 3.10 Admittance phase angle in parallel GLC branch
According to the definition of characteristic impedance, the values of L and C may be
expressed as
0 0
1, L C
ρω ω
= = (3.32)
and the susceptance may be expressed as a function of characteristic impedance
0
0
1 1B C
L
ω ωω
ω ω ω ρ
= − = −
(3.33)
Similarly as for parallel branch the relative deviation from resonant frequency
0 0
0 0
f f
f f
ω ωδ
ω ω= − = − (3.34)
26
and
Qξ δ= (3.35)
A quite interesting and unusual situation may be present in a parallel circuit with two
branches RL and RC (Figure 3.11). Currents at resonant frequency are shown in Figure 3.12
where the compensation of reactive currents takes place.
Figure 3.11 Circuit with parallel braches RL and RC
Figure 3.12 Currents in a parallel RL and RC circuit at resonance
The compensation condition is given by
1 1 2 2sin sin 0I Iϕ ϕ+ = (3.36)
also
1 21 2 2 2
1 2
0X X
B BZ Z
+ = = − − (3.37)
Applying (3.37) to the circuit in Figure 3.11 we obtain
2 2 2
21
2 2 2
1
01
L C
R L RC
ω ωω
ω
− + =+ +
(3.38)
rewriting this we get
2 2 22 1
1L LL R R
C C Cω − = −
(3.39)
where /L C is equal to the squared value of characteristic impedance.
If the resistances of both branches are equal to characteristic impedance
27
1 2
LR R
C= = (3.40)
then by every frequency the resonance of currents appears. This is a rare example of a
frequency insensive circuit. In this particular case, the circuit impedance is
1
1
L Lj L j
C C C LZ
CL Lj L j
C C C
ωω
ωω
+ −
= =
+ + −
(3.41)
also the impedance is purely resistive and equal to the characteristic impedance.
For two different resistances the resonant frequency is given as
21
022
1'
LR
CLL RCC
ω−
=−
(3.42)
If the student is interested in the analysis of more complicated cases it is strongly advisable
to refer to [3].
3.4. Literature
[1] John Bird, “Electrical Circuit Theory and Technology”, Elsevier, Linacre House, Jordan
Hill, Oxford OX2 8DP, UK, 2007, pp. 349-368
[2] Tadeusz Łobos, Jacek Rezmer, Przemysław Janik, Zbigniew Wacławek, “Prony and
nonlinear regression methods used for determination of transient parameters in wind
energy conversion system”, IEEE Lausanne PowerTech, Switzerland, 1-5 July 2007, 6 pages
[3] Tadeusz Cholewicki, “Elektrotechnika Teoretyczna”, WNT, Warszawa 1973 pp. 391-
422
[4] Stanisław Bolkowski, „Teoria obwodów elektrycznych”, WNT, Warszawa, 1995 pp.
150-168
4. Analog filter design methodology for harmonic cancelation
4.1. Problem formulation
One of the various sources of harmonics are widespread lightening systems. Modern lighting
systems - street lighting, interior lighting, illumination of buildings must meet multi-criteria
28
demands imposed by standards and users. One of the demands is reduction of harmonics
and interharmonics and guaranteeing good power quality.
Traditional and compact fluorescent lamps, powered via a stabilizing-ignition system, sodium
and mercury discharge lamps, light sources powered via power converters contribute
significantly to improving the lighting quality and reducing energy costs. Unfortunately, the
common feature of all these devices is the non-linearity [1]. The disadvantageous feature is
strengthened by other nonlinear devices powered from the same pcc .
Main feature of the nonlinear load is strongly non-sinusoidal current drawn from the grid.
This current causes non-sinusoidal voltage drops on transformers’ impedances and power
lines. It distorts the voltage at pcc, to which other users are connected. Nonlinear loads are
the source of power quality deterioration [2], [3].
Non-sinusoidal voltages and currents cause technical problems and obstruct the function of
network components (e.g. e automation and protection units), additionally negatively
influence other loads [4].
Reduction of harmonics in distribution systems is forced as a result of technical reasoning.
Also legal regulations imposed on distribution companies and manufacturers of electrical
equipment determine acceptable levels of voltage and current distortion [5], [6], [7].
There is a need for a simple and effective way of reducing the content of harmonics and
interharmonics. Analog filters [8] are in use, but a detailed and tedious design process for
every application side is required. This article presents the theoretical basis for an efficient
filter design, together with sample calculations and simulations carried out for selected
characteristic harmonics. The aim of the authors was to deliver an accurate and easy tool do
design filters distinguishing selected harmonics and interharmonics. Therefore, the
implementation of filters may be very precise and effective. Without detailed calculation for
each installation side, application of “average” filter is not advisable.
29
4.2. Higher order Harmonics originating in nonlinear loads
Nonlinear lighting circuits draw periodical, non-sinusoidal currents i. Generally, periodic
current can be expressed as the sum of sine waves, forming a Fourier series [8]
( ) ( ) ( )0 m1 0 1 mk k k
k 1
i t I t I tω ψ ω ψ∞
>
= + + +∑sin sin (4.1)
where Imk means the amplitude of the k-th harmonic component, ωk is angular velocity and
ψk initial phase shows a signal distorted by 5,7,9 and 11 harmonics.
0 0.01 0.02 0.03 0.04 0.05 0.06 -1.5
-1
-0.5
0
0.5
1
1.5
time[s]
I [pu]
Figure 4.1 Distorted current signal
Harmonics have a direct impact on cables, especially neutral, and transformers. Integer
multiples of third harmonic add up in the neutral. Current in which may be, in an extreme
case, up to three times higher than in the phase wire. The presence of higher harmonics in
the current causes additional heating of cables, therefore, it is dangerous for the insulation.
The presence of higher harmonics in voltage increases the current drawn by the capacitors,
since the capacitor impedance decreases with increasing frequency.
The magnetic flux in the transformer originating from the third current harmonic is closed
outside the core, resulting in the formation of eddy currents in other parts of the
transformer [9]. Additional energy losses heat up the transformer interior, so accelerate the
aging process of isolation [2].
Higher harmonics also have an adverse impact on the protection [4], in particular the
operation of differential-current protection. Harmonics cause also false measuring of
electricity meters. They are dangerous for the insulation of cables and compensating
capacitor banks, causing heating and damage to insulation. Harmonics cause resonances
which are difficult to predict and result in damage to measuring transformers [3].
Higher voltage harmonics are especially dangerous for drives [2]. They generate additional
heat loss together with fluctuating value of torque on the shaft.
30
4.3. Currents drawn by a non-sinusoidal load, general characteristic
Consider a system consisting of a non-linear receiver, connected to the grid (Figure 4.2, left),
which in simple terms can be represented in terms of the actual source of sinusoidal voltage
(Figure 4.2, right).
Figure 4.2 non-linear load connected to the grid (l), representation of the network with voltage source (r)
We limit the discussion to steady state. Then, even assuming sinusoidal voltage, the load
draws deformed, nonlinear current, containing higher harmonics. Due to the voltage drop
across the impedance Zg (Fig. 2), the load will be fed by non-sinusoidal voltage anyway.
The complete elimination of higher harmonics from the currents and voltages waveforms is
not possible. We can only minimize the distortion, separating the distortion sources from the
grid. Distorted waveforms will be “closed” in a certain “loop”, and load is powered by the
fundamental harmonic only. Separators are subsystems with appropriately selected
frequency characteristics, such as resonant circuits or filters.
Supposing, that a non-linear load drew the current given by (4.1). Actually, other harmonics
may be present due to preexisting voltage distortion. An effective way to determine
harmonic (interharmonic) content in the current is direct measurement of the spectrum.
Starting from the compensating theorem [8],[10] and (4.1) we can replace the load by a
battery of current sources connected in parallel, each corresponding to the specific
harmonic current (Figure 4.3).
31
A
B
u(t)e(t)
( )gZ jω i0(t)
i1 1k
i(k
i
Figure 4.3 Load represented as a battery of current sources
If we place ideal serial resonant branches between the source and the load (Figure 4.4),
tuned respectively to each of the higher harmonics, the currents of those frequencies will be
completely "separated" from the grid, and the current drawn from the system will contain
only the component of fundamental frequency
( ) ( )m1 0 1i t I tω ψ′ ′ ′= +sin (4.2)
A
B
e(t)
( )gZ jω
L2
C2
L3
C3
L(
C(
1i′
( )0i t′( )i t′
( )u t′1ki′
(ki′
Figure 4.4 Load with ideal resonant branches
However, application of resonant circuits with a very high quality factor is dangerous due to
the possibility of significant over-voltages or over-currents. Therefore, instead of ideal
branches, circuits in which there is a possibility of regulating the quality factor, are in use.
We can modify the circuit in Figure 4.4 by adding a resistor in parallel to the inductance
(Figure 4.5). The quality factor can be changed varying the resistance.
A
B
e(t)
( )gZ jω
L2
C2
L3
C3
L(
C(
R2
R3 R
(
( )0i t′( )i t′
1i′( )u t′
1ki′
(ki′
32
Figure 4.5 Load with resonant branches including resistor
To make a bit more in-depth analysis of the system in Figure 4.5, suppose that this circuit,
except the load is a linear circuit, for which the principle of superposition applies. On this
basis, the circuit can be studied individually for each harmonic. An example of such a system
for the k-th harmonic is shown in Figure 4.6.
Lk
Rk
Ck
uk
Ls
Rs
B
A
ki′RLku
Cku
Cki
RkiLk
i
ki
sku
( )kZ jω′
Figure 4.6 Circuit for the k-th harmonic reduction
Assigning the impedance of the k-th resonance circuit by Zk(jωk) and a summary of all other
impedances for the rest of harmonics by Z'k(jωk) we can define the current of the k-th
harmonic as
( ) ( )
( ) ( ) ( )k k k k
k k
k k k k s k
Z j Z jI I
Z j Z j Z j
ω ωω ω ω
′ ′=′ +
/ /
/ / (4.3)
where
( ) ( ) ( ) ( )( ) ( )k k k k
k k k kk k k k
Z j Z jZ j Z j
Z j Z j
ω ωω ω
ω ω
′′ =
′+/ / (4.4)
If we select the impedances of resonance circuit in such a way that
( ) ( )k k k kZ j Z jω ω′ (4.5)
then, the complex value of the k-th current harmonic is given by
( )
( ) ( )k k
k k
k k s k
Z jI I
Z j Z j
ωω ω
′≈+
(4.6)
We can therefore conclude that for absolute values of impedances Zk(jωk) small enough at
resonance frequencies, then the harmonic content in current and supply voltage u(t), will be
significantly reduced, in comparison to the circuit in Fig. 3. A brief description of the basic
33
concepts necessary to analyze resonant circuits is given the special respect to the design
method presented here.
4.4. Basic dependencies in the resonant circuit selected for the filter
The basic feature describing a resonant circuits is its impedance, defined as a complex
function of a real input variable, mostly frequency. Using the impedance function we can
compute important parameters - resonance frequency, impedance at resonance, bandwidth.
Let us consider a circuit shown in Figure 4.7.
Lk
Rk
Ck
u
ik
RLku
Cku
Cki
RkiLk
i
( )kZ jω
U
RLkU
RkI
CkkI I=
LkI
CkU
Figure 4.7 Resonance branch and corresponding vector diagram at phase resonance
Assuming that the vectors in Figure 4.7 represent sinusoidal functions with the pulse ω, the
complex impedance is defined as the ratio of complex voltage to the complex current value
( )2 2 2
k k k k k kk 2 2 2 2 2 2
k k k k kk k k k
R j L R L R LU 1 1Z j j
I R j L j C CR L R L
ω ω ωω
ω ω ωω ω
= = + = + −
+ + + (4.7)
Hence, considering the phase resonance condition
( )2k k
k 2 2 2kk k
R L 1Z j 0
CR L
ωω
ωω= − =
+Im (4.8)
We compute the frequency f (angular frequency ω) at resonance
( ) ( )k k
k k
k k2 2
L L
k k k kR R
1 1f
L C 2 L C
ωπ
= =
− −
, (4.9)
using the computed value of ωk we can compute the impedance at resonance
( )2
Lkk k Rk
2Lk
k k Rk
21k k
2 2 L Ck k k k
k k k k0 2 2 2 2 21k kk k k k k
L C
R L
R L LZ j Z
R CR L R L
ωω ω ω
ω
−
−
= ⇒ = = = =+ +
(4.10)
34
Quality factor Q is an important feature characterizing a resonant circuit. It is defined as a
ratio of the maximal energy of the electromagnetic field and energy radiated as heat at
resonance during one period [8].
0 0
WQ 2
T Pπ= max (4.11)
Where Wmax - maximal energy of the electromagnetic field, P0 – average power radiated
during one period, T0 – period of the waveform under consideration. For the circuit in Figure
4.7 the expression for the quality factor is given by
k k
1C k k
k kk0 k k k
R CQ R
Z L L
ω
ω= = = (4.12)
The quality factor is followed by another useful parameter – transmission bandwidth.
Transforming ((4.7)) into ((4.13)) we obtain corresponding plot of the impedance absolute
value near the resonant frequency (Figure 4.8)
( )( ) ( )
( ) ( )
2 22 2k k k k
k 2 2 2k k k
R 1 L C LZ j
C R L
ω ωω
ω ω
− +=
+
(4.13)
130 135 140 145 150 155
0
4
8
12
16
20
24
28
32
36
40
∆fk
f Hz
fk1 f
k2
fk
Figure 4.8 Computation of the transmission bandwidth
We observe (Figure 4.8) that the transmission bandwidth is given as the difference of
frequencies fk1 and fk2 for which the impedance absolute value is 2 higher than at
resonance (power reduced by half).
35
Exact expression for transmission bandwidth can be given only for relatively simple circuits.
In the presented case we apply approximated approach describing the impedance as a
function of fractional deviation from the resonant frequency
k k
fx
f
ω
ω= = (4.14)
Putting (4.14) into (4.7) and using (4.9), (4.10) and (4.12) we get a new expression for the
impedance
( )( ) ( )
3222
k2 2k k0k 2 2 2
k k
Q 1Q ZZ x x j x 1
Q x 1 Q x
−
= + − + −
(4.15)
Usually, Q factor is significantly higher than 1. Assuming analysis around resonance
frequency, we can say that
k
Q 1
x 1≈
(4.16)
then
( ) ( )k k0 kZ x Z 1 j2Q x 1 ≅ + − (4.17)
and
( ) ( )2 22 2k k0 kZ x Z 1 4Q x 1 ≅ + −
(4.18)
defining
k1 k21 2
k k
f fx x
f f= =, (4.19)
we obtain
( )22k 1 21 4Q x 1 2+ − =, (4.20)
therefore
1 2k
1x 1
2Q= m, (4.21)
Finally, the bandwidth is
( )k k2 k1 2 1 k kk
1f f f x x f f
Q∆ = − = − = (4.22)
36
In (4.23) we recognize a plain relationship between Q factor and bandwidth. Quality factor Q
may be determined using the resonance characteristic
kk
k
fQ
f∆= (4.23)
The influence of various Q factor values obtained by (4.24) on the shape of the resonance
curve is shown in Figure 4.9, where k0Z 10 Ω= , , , ,kQ 5 10 20 50= .
( )( ) ( )
322 2k4 2k k0
k 2 2 4 2k k
Q 1Q ZZ x x x 1
Q x 1 Q x
−= + −
+ − (4.24)
0 0.5 1 1.5 2
0
200
400
600
x
Qk=10
Qk=20
Qk=50
Qk= 5
Figure 4.9 Impedance vs. Q factor
Unfortunately, there is a discrepancy between the minimum of impedance and resonance
frequency. In practice, it may have a negative impact on filtering accuracy. Therefore, it is
advisable to apply a corrective value given by (4.25). It was derived using the small
parameter method.
2 4k k
1 1k0 k
Q 2Qf f 1≅ − + (4.25)
4.5. Proposed design for efficient harmonic filtration
Figure 4.10 shows a simplified model of a filter for arbitrary selected harmonics or
interharmonics. The idea may be expanded on every frequency, not only an integer multiply
of fundamental
37
Figure 4.10 Filter for selected harmonics
Using the method given above, the set of expression enabling fast and accurate
determination of RLC parameters will be presented.
Usually, there are required values for resonance frequency fk, impedance at resonance Zk0,
and Q factor (or bandwidth Δfk). For Q factor smaller than 10, it is recommended to apply
the correction given by (4.25). Then, fk is given by
2 4k k
k0k
1 1
Q 2Q
ff
1≅
− + (4.26)
From (4.10) and (4.12) we get
2
k k k0R Q Z= (4.27)
Then we compute the wave impedance ρ
22 2 2k k
k k02k k
L RQ Z
C Qρ = = = (4.28)
From(4.28) and (4.9) we obtain
2k
k k02
k k
Q1L Z
2 f Q 1π=
− (4.29)
and
k kk 2 2 2 2
k k0k k k0 k
L L 1 1 1C
2 f ZQ Z Q 1πρ= = =
− (4.30)
4.6. Computation of filtration parameters
Mathematical expressions from previous chapters were used in realization of a four section
filter for a priori chosen harmonics
03 05
07 11
f 150Hz f 250 Hz
f 350 Hz f 550 Hz
= =
= =
, ,
, (4.31)
38
The absolute value of impedance for each resonance frequency can be chosen
independently. Equal value was assumed for every harmonics in order to simplify the
presentation
03 05 07 011Z Z Z Z 14 Ω= = = = (4.32)
Similarly, the Q factor for every component filter was equalized. Four different cases are
summarized in Table 4-1.
Table 4-1 Various values of the Q factor
Case No. 1 2 3 4
= = =3 5 7 11
Q Q Q Q 5 10 20 50
Tables show the RLC and f values for Q factors from Table 4-1.
Table 4-2 RLC values for the case 1, Q=5
k = 3 k = 5 k = 7 k = 11
R 350Ω 350Ω 350Ω 350Ω
L 0.074H 0.044H 0.032H 0.020H
C 15.16µF 9.10µF 6.65µF 4.135µF
f 153.03Hz 255.05Hz 357.07Hz 561.11Hz
0 100 200 300 400 500 600 700 800 900 1000 1100
10
100
1000
k=5
k=3
k=7
k=11
f Hz
Figure 4.11 Absolute value of impedance for Q=5
39
R 1400Ω 1400Ω 1400Ω 1400Ω
L 0.148H 0.89H 0.637H 0.040H
C 7.579µF 4.547µF 3.248µF 2.067µF
f 150.75Hz 251.25Hz 351.75Hz 552.76Hz
0 100 200 300 400 500 600 700 800 900 1000 1100
10
100
1000
k=5
k=3
k=7
k=11
f Hz
Figure 4.12 Absolute value of impedance for Q=10
Table 4-4 RLC values for the case 1, Q=20
k = 3 k = 5 k = 7 k = 11
R 1400Ω 1400Ω 1400Ω 1400Ω
L 0.148H 0.89H 0.637H 0.040H
C 7.579µF 4.547µF 3.248µF 2.067µF
f 150.75Hz 251.25Hz 351.75Hz 552.76Hz
40
Table 4-3 RLC values for the case 1, Q=10
k = 3 k = 5 k = 7 k = 11
0 100 200 300 400 500 600 700 800 900 1000 1100
10
100
1000 k=5
k=3
k=7
k=11
f Hz
Figure 4.13 Absolute value of impedance for Q=20
Table 4-5 RLC values for the case 1, Q=50
k = 3 k = 5 k = 7 k = 11
R 35kΩ 35kΩ 35kΩ 35kΩ
L 0.743H 0.446H 0.318H 0.203H
C 1.516µF 0.909µF 0.650µF 0.413µF
f 150.03Hz 250.05Hz 350.07Hz 550.11Hz
0 100 200 300 400 500 600 700 800 900 1000 1100
10
100
1000
10000
k=5k=3
k=7
k=11
f Hz
Figure 4.14 Absolute value of impedance for Q=50
4.7. Simulation of nonlinear load working with a filter
Simulation is an effective way to test the presented approach to filter design. The filter was
implemented in Power System Bolockset [11], a part of Matlab [12] and shown in Figure
4.15.
41
io (t)
+i-
i o
ih7(t)
+i -
i h7
ih5(t)
+i -
i h5
ih3(t)
+i -
i h3
ih11(t)
+i -
i h11
i7i5i3 i11i1
i (t)
+i-
i
e (t)
Rg Lg
R7 L7R5 L5R3 L3 R11 L11
C7C5C3 C11
Figure 4.15 Simulated one phase system
To show clearly the filtration effect, the presence of 3, 5, 7, 9, 11 current harmonic was
assumed. Every harmonic with an amplitude of 20% of the fundamental. Due to a significant
difference between the impedance of the network Zg and the impedance Zf of the filter (50
to 14) the effect of filtering is very clear (Fig. 15). The module of impedance Zf should be
selected depending on the impedance of the network, the expected effect of filtration and
maximum overcurrent and overvoltage in the filter.
Figure 4.16 Current drawn by nonlinear load
42
Figure 4.17 Current drawn from the pcc
Figure 4.18 Current in the filter branch for 3-th harmonic
0 0.01 0.02 0.03 0.04-4
-3
-2
-1
0
1
2
3
4
time [s]
Figure 4.19 Current in the filter branch for 11-th harmonic
43
Figure 4.18 and Figure 4.19 shows currents flowing in the branches of the filter, designed for
3 and 11 harmonics. These graphs show the effectiveness of filtration.
An effective way to eliminate harmonics is an appropriate use of analog filters. Selecting an
adequate filter is primarily reduced to the analysis of the load structure and harmonics levels
in the current. The next step is to find a filter structure with appropriate parameters.
4.8. Literature
[1] F. C. de la Rosa, Harmonics and power systems (Boca Raton FL:, CRC Press 2006)
[2] R. Dugan, Electrical Power Systems Quality, (McGraw-Hill, New York 1996)
[3] J. Arrilaga, N. R. Watson, Power System Quality Assessment (Chihester England, John
Wiley & Sons, 2000)
[4] F. Wang, M. H. J. Bollen, Quantifying the potential impacts of disturbances on power
system protection, Proceednings of ‘Developnents in Power System Protection Conference’,
Amsterdam 9-12 April, 2002, 262-266
[5] EN 50160, Standard, Voltage Characteristics In Public Distribution Systems
[6] PN-EN 61000-3-2, Standard, Limits for harmonic current emmissions (equipment input
current up to and including 16A per phase).
[7] IEC 662, Standard, Sodium Vapor Lamps
[8] J.Bird, Electric circuit theory and technology (London, Elsevier, 2007)
[9] MGM Transformer online: www.mgm-transformer.com
[10] U. A. Bakshi, A. V. Bakshi, Circuit Theory (Shanivar Pet, Pune, 2009)
[11] The Matworks: Power System Blockset. User’s Guide, (The Math Works Inc. 2000)
[12] The Matwroks: Matlab. The Language of Technical Computing. Using Matlab, (The
Math Works Inc. 2001)
5. Fourier Techniques for Spectral Analysis
Harmonics have always been present in power systems. Recently, due to the widespread use
of a power electronics system resulting in an increase in their magnitude, they have become
a key issue in installations [1]. A harmonic distortion can be considered as a sort of pollution
of the electric system which can cause problems if the sum of harmonics currents exceeds
certain limits [1].
44
Fourier transform is one of the most powerful tools in a spectral analysis. The assessment of
harmonics content in currents and voltages is usually done with Fourier Transform. In this
chapter, the outline of the discrete Fourier Transform will be given, and is based mainly on
[2].
5.1. Discrete Time Fourier Transform Definition
Let us firstly formulate and explain the definition.
The Fourier Transform of a discrete series DTFT ( )jX e ω of the signal [ ]x n is defined as
( ) [ ]j j n
n
X e x n eω ω∞
−
=−∞
= ∑ (5.1)
Generally, ( )jX e ω is a complex function of a real variable ω and may be written as a sum of
real and imaginary part
( ) ( ) ( )j j j
re imX e X e jX eω ω ω= + (5.2)
( ) ( )j j
re imX e and X eω ω are the real and imaginary parts of ( )jX e ω
and consequently real functions of the independent variable ω .
Sometimes, the exponential form of DTFT is more suitable and readable
( ) ( ) ( )jj jX e X e eθ ωω ω= (5.3)
where ( )θ ω
( ) ( ) arg jX e ωθ ω = (5.4)
and returns directly the amplitude spectrum and phase spectrum. Therefore,
( )jX e ωis the absolute value of the function or amplitude spectrum and
( )θ ω is the phase as a function of frequency or briefly phase spectrum. Both quantities are
functions of real variable ω .
Assuming that [ ]x n is a real function (almost all signals in the power quality area),
characteristic features can be noticed
• ( ) ( )j j
reX e and X eω ω are even functions of ω
• ( ) ( )j
imand X e ωθ ω are odd functions of ω .
Both are a consequence of complex exponential function properties, compactly written as
45
( ) ( ) ( ) ( ) ( )2j k jj j jX e X e e X e eθ ω π θ ωω ω ω+= = (5.5)
for any integer k .
Computation of Fourier Transform of a power sequence describing a causal system has been
chosen to illustrate practical use of the definition of DTFT.
The signal, e.g. impulse response of the mentioned system
[ ] [ ]1nx n a n= ⋅ (5.6)
where 1a < to guarantee convergence.
DTFT is given as
( ) [ ] ( )0 0
11
1
nj n j n n j n j
jn n n
X e a n e a e aeae
ω ω ω ωω
∞ ∞ ∞− − −
−=−∞ = =
= ⋅ = = =−∑ ∑ ∑ (5.7)
and again 1jae aω− = < for convergence.
Amplitude and phase spectrum of DTFT (5.8) for 0.5a = may be presented graphically.
( ) 1
1 0.5
j
jX e
e
ωω−
=−
(5.8)
Figure 5.1 shows the amplitude spectrum of (5.8) and Figure 5.2 shows the corresponding
phase spectrum.
Figure 5.1 Amplitude spectrum
Figure 5.2 Phase spectrum
46
The two pictures above indicate two important general features of the transform:
• DTFT of the sequence [ ]x n is a continuous function of frequency ω .
• DTFT of the sequence [ ]x n is a periodic function of ω with the period 2π .
5.2. Discrete Fourier Transform Definition
In practice the Discrete Fourier Transform DFT, an alternation of DTFT is in use. The
definition and examples will be given below.
The relationship between the finite sequence [ ]x n (bounded in its length) 0 1n (≤ ≤ − and
its DTFT ( )jX e ω may be obtained through equidistant sampling of the function ( )jX e ω on
the ω axes in the interval 0 2ω π≤ ≤ at points 2 / , 0 1k k ( k (ω π= ≤ ≤ − .
So, the defining equation may be written
[ ] ( ) [ ]1
2 /
02 /
(j j kn (
n
X k X e x n ek (
ω π
ω π
−−
=
= == ∑ (5.9)
where 0 1k (≤ ≤ − .
is a complex series of the length ( in frequency domain.
Discrete series [ ]X k is called a Discrete Fourier Transform (DTF) of the sequence [ ]x n .
Introducing the supposition 2 /j (
(W e π−= the DFT may be given in frequently used form
[ ] [ ]1
0
(kn
(
n
X k x n W−
=
=∑ (5.10)
where 0 1k (≤ ≤ − .
There is a way to invert the transformation and return to the finite time series. The
procedure, called Inverse Discrete Fourier Transform IDFT, is given by
[ ] [ ]1
0
1 (kn
(
k
x k X k W(
−−
=
= ∑ (5.11)
where 0 1n (≤ ≤ − .
An example illustrates the practical use of the DFT definition.
n
X[n]
1
0 1 2 N-1
47
Figure 5.3 Discrete signal of the length N.
The DFT of the signal in Figure 5.3 contains also N points in the frequency domain
[ ] [ ] [ ]1
0
0
0 1(
kn
( (
n
X k x n W x W−
=
= = =∑ (5.12)
where 0 1k (≤ ≤ − . All points have the same value, one.
Another example considers a signal given by
[ ]1,
0, 0 1, 1 1
n mx n
n m m n (
==
≤ ≤ − + ≤ ≤ − (5.13)
the corresponding DFT
[ ] [ ] [ ]1
0
, 0 1(
kn km km
( ( (
n
X k x n W x m W W k (−
=
= = = ≤ ≤ −∑ (5.14)
where 0 1k (≤ ≤ − .
It is very convenient for the practical numerical computation to present the DFT equation in
a matrix form. DFT given as [ ] [ ]1
0
(kn
(
n
X k x n W−
=
=∑ may be compactly written in the matrix
form
(=X D x (5.15)
where
[ ] [ ] [ ]0 1 1T
X X X ( = − X K (5.16)
represents the signal in frequency domain,
[ ] [ ] [ ]0 1 1T
x x x ( = − x K (5.17)
is the signal in time domain to be transformed and
( )
( )
( ) ( ) ( )2
11 2
2 12 4
1 2 1 1
1 1 1 1
1
1
1
(
( ( (
(
( ( ( (
( ( (
( ( (
W W W
W W W
W W W
−
−
− − −
=
D
L
L
L
M M M O M
L
(5.18)
is the transformation matrix.
Also the Inverse Discrete Fourier Transform defined by [ ] [ ]1
0
1 (kn
(
k
x n X k W(
−−
=
= ∑ may be
written in a matrix form
48
11(
(
−=x D X (5.19)
where
( )
( )
( ) ( ) ( )2
11 2
2 11 2 4
1 2 1 1
1 1 1 1
1
1
1
(
( ( (
(
( ( ( (
( ( (
( ( (
W W W
W W W
W W W
− −− −
− −− − −
− − − − − −
=
D
L
L
L
M M M O M
L
(5.20)
is the inverse transformation matrix.
The transformation matrix and inverse transformation matrix are joined by the expression
1 *1( (
(
− =D D (5.21)
5.3. Discrete Fourier Transform expressed using trigonometric functions
The well known Euler’s formula (5.22) allows us to rewrite the equations for DFT
( ) ( )cos sinj
e jφ φ φ− = − (5.22)
into the trigonometric form
[ ] [ ] ( ) ( )1
0
cos 2 / sin 2 /(
n
X k x n nk ( j nk (π π−
=
= − ∑ (5.23)
where
[ ]X k is the k-th component of DFT, meaning [ ] [ ] [ ]0 , 1 ,..., 1X X X ( − ,
k is the index of samples of DFT in frequency domain, meaning 0, 1, 2, 3,..., 1k (= − ,
[ ]x n is the sequence of input samples, meaning [ ] [ ] [ ]0 , 1 ,..., 1x x x ( − ,
n is the index of input samples in time domain, meaning 0, 1, 2,3,..., 1n (= − ,
( is the number of samples of the input sequence and the number of frequency points in
the DFT sequence.
Indices n for the input samples and k for the samples of DFT always increase from 0 to
1( − in the standard notation of DFT. That means, for ( input samples in time domain,
DFT computes the spectrum content of the input signal at ( equidistant points of the
frequency axes.
That means, for ( input samples in time domain, DFT computes the spectrum content of
the input signal at ( equidistant points of the frequency axes.
49
The value of ( is therefore an important parameter of DFT, with a direct impact on:
- the number of required input samples
- the spectral resolution in frequency domain
- the time needed for processing a signal of length N to the DFT representation
To illustrate the practical application of DFT some examples are sown.
Example I.
The detailed computation of four values in frequency domain will be shown.
If 4( = , then both n , and k k increase from 0 to 3. The particular expression for DFT has
the form:
[ ] [ ] ( ) ( )3
0
cos 2 / 4 sin 2 / 4X k x n nk j nkπ π = − ∑ (5.24)
Writing all spectral terms of DFT for the first sample, 0k = , we obtain
[ ]( ) ( ) ( ) ( )( ) ( ) ( ) ( )( ) ( ) ( ) ( )( ) ( ) ( ) ( )
0
0 cos 2 0 0 / 4 0 sin 2 0 0 / 4
1 cos 2 1 0 / 4 1 sin 2 1 0 / 4
2 cos 2 2 0 / 4 2 sin 2 2 0 / 4
3 cos 2 3 0 / 4 3 sin 2 3 0 / 4
X
x jx
x jx
x jx
x jx
π π
π π
π π
π π
=
⋅ ⋅ − ⋅ ⋅ +
⋅ ⋅ − ⋅ ⋅ +
⋅ ⋅ − ⋅ ⋅ +
⋅ ⋅ − ⋅ ⋅
(5.25)
For second sample of DFT, 1k = the equation is
[ ]( ) ( ) ( ) ( )( ) ( ) ( ) ( )( ) ( ) ( ) ( )( ) ( ) ( ) ( )
1
0 cos 2 0 1/ 4 0 sin 2 0 1/ 4
1 cos 2 1 1/ 4 1 sin 2 1 1/ 4
2 cos 2 2 1/ 4 2 sin 2 2 1/ 4
3 cos 2 3 1/ 4 3 sin 2 3 1/ 4
X
x jx
x jx
x jx
x jx
π π
π π
π π
π π
=
⋅ ⋅ − ⋅ ⋅ +
⋅ ⋅ − ⋅ ⋅ +
⋅ ⋅ − ⋅ ⋅ +
⋅ ⋅ − ⋅ ⋅
(5.26)
The third sample is obtained in the same way, 2k =
[ ]( ) ( ) ( ) ( )( ) ( ) ( ) ( )( ) ( ) ( ) ( )( ) ( ) ( ) ( )
2
0 cos 2 0 2 / 4 0 sin 2 0 2 / 4
1 cos 2 1 2 / 4 1 sin 2 1 2 / 4
2 cos 2 2 2 / 4 2 sin 2 2 2 / 4
3 cos 2 3 2 / 4 3 sin 2 3 2 / 4
X
x jx
x jx
x jx
x jx
π π
π π
π π
π π
=
⋅ ⋅ − ⋅ ⋅ +
⋅ ⋅ − ⋅ ⋅ +
⋅ ⋅ − ⋅ ⋅ +
⋅ ⋅ − ⋅ ⋅
(5.27)
The final one in this example, 3k =
50
[ ]( ) ( ) ( ) ( )( ) ( ) ( ) ( )( ) ( ) ( ) ( )( ) ( ) ( ) ( )
3
0 cos 2 0 3 / 4 0 sin 2 0 3 / 4
1 cos 2 1 3 / 4 1 sin 2 1 3 / 4
2 cos 2 2 3 / 4 2 sin 2 2 3 / 4
3 cos 2 3 3 / 4 3 sin 2 3 3 / 4
X
x jx
x jx
x jx
x jx
π π
π π
π π
π π
=
⋅ ⋅ − ⋅ ⋅ +
⋅ ⋅ − ⋅ ⋅ +
⋅ ⋅ − ⋅ ⋅ +
⋅ ⋅ − ⋅ ⋅
(5.28)
Looking at the results above we can formulate an important remark:
Every sample of [ ]X k DFT is a “point after point” summation of a product term consisting
of subsequent values of the input series and a complex term ( ) ( )cos sinjφ φ− .
We see in that example that the frequencies of sinusoidal components depend on the
sampling frequency pf , and the number of samples ( .
Example II.
If we sample an analog signal with the ratio of 500 samples per second and compute
afterward a 16-point DFT from these samples, then the fundamental frequency of the sinus
waves to which we correlate the signal to be transformed is
/ 500 /16 31, 25Hzpf ( = = (5.29)
Computed values of ( )X k DFT called “stripes”, are determined at points being exact
multiplies of the fundamental frequency. In presented case, the specific values are
[ ]0X - 1. strip with analysis frequency = 0x31,25 = 0 Hz.
[ ]1X - 2. strip with analysis frequency = 1x31,25 = 31,25 Hz,
[ ]2X - 3. strip with analysis frequency = 2x31,25=62,5 Hz,
[ ]3X - 4. strip with analysis frequency = 3x31,25 = 93,75 Hz.
.....................
[ ]15X - 16. strip with analysis frequency 15x31,25=468.75 Hz.
The values of ( subsequent points on the frequency axes are given by
p
a
mff
(= (5.30)
The resolution of DFT is given as ratio of the sampling frequency to the number of samples
pf
f(
∆ = (5.31)
51
Instantaneous power of the signal [ ]pX k is expressed as a square of the absolute value of
the spectrum.
( ) ( ) ( ) ( )2 2 2
P re imX k X k X k X k= = + (5.32)
Example III.
An input signal ( )x t should be sampled. The signal contains two spectral components with
the frequencies 1 kHz and 2 kHz, and is given as
( ) ( ) ( )312 4
sin 2 1000 sin 2 2000x t t t ππ π= ⋅ ⋅ + ⋅ ⋅ + (5.33)
Afterwards, 8-point DFT of the sampled signal should be computed.
The approach to the problem may be the following:
Choosing a sampling frequency pf , we get equidistant samples every 1/p pT f= . We need
exactly 8 input samples, allowing the computation of 8 - points DFT. Eight elements of the
series [ ]x n are equal to ( )x t sampled at the time instants pnT
[ ] ( ) ( ) ( )312 4
sin 2 1000 sin 2 2000p p px n x nT nT nT ππ π= = ⋅ ⋅ + ⋅ ⋅ + (5.34)
If we choose the sampling frequency 8000pf = samples/second, then DFT determines
spectral components of the signal [ ]x n at certain points of the frequency axes pfk
(⋅ , which
is at 0 kHz, 1 kHz, 2 kHz ... 7 kHz.
Numerical values [ ]x n of the sampled signal given by (5.34) are: x[0]= 0,3535, x[1]= 0,3535,
x[2]= 0,6464, x[3] = 1,0607,x[4]= 0,3535, x[5]=-1,0607, x[6] = -1,3535, x[7] = -0,3535.
These values of samples [ ]x n are shown as dots on the continuous line ( )x t on figures
below illustrating the DFT.
Let us now compute the DFT values.
For 1k = , or the DFT component with the frequency 1 kHz, 8000
8
pkf
(= , the equation in this
case has the form:
( ) [ ] [ ]7
0
2 21 cos sin
8 8n
n nX x n jx n
π π
=
= −
∑ (5.35)
Substituting the values of samples x[n] into the equation and putting the cos terms into the
left column and the sin terms into the right column we obtain
52
X[1]= 0,3535 · 1,0 -j (0,3535 · 0,0)+ for n=0
+ 0,3535 · 0,707 -j(0,3535 · 0,707)+ for n=1
+ 0,6464 · 0,0 -j(0,6464 · 1,0)+ for n=2
+ 1,0607 · -0,707 -j(1,0607 · 0,707)+ for n=3
+0,3535 · -1,0 - j(0,3535 · 0,0)+ for n=4
-1,0607 · -0,707 -j(-1,0607 · -0,707)+ for n=5
-1,3535 · -0,0 -j(-1,3535 · -1,0)+ for n=6
-0,3535 · 0,707 -j(0,3535 · -0,707)= for n=7
= 0,3535 + 0
+ 0,250 -j0,250 +
0 -j0,6464 +
- 0,750 -j0,750+
-0,3535+ 0
+0,750 -j0,750+
0 -j1,3535
-0,250= +j0,250=
X[1]= 0—j4,0 904 je−=
o
It should be noticed, that the sin and cos lines used in the analysis have the frequency of
1kHz (Figure 5.4), which is one period in the sampling interval.
Figure 5.4 Signal x(t) and sin, cos curves with 1 kHz used for sampling
The computational results shows that the input signal [ ]x n contains a component with the
frequency 1kHz. Detailed information of the component for 1k = :
0 1 2 3 4 5 6 -1.5
-1
-0.5
0
0.5
1
1.5
k=1
53
Absolute value of the first component ( )1X
( )1 4X = (5.36)
Phase
( )( )( )
01
1 901
imag
real
XX arctg
Xφ
= = −
(5.37)
Instantaneous power
( ) ( )2
1 1 16PX X= = (5.38)
Let us now compute the second component for 2k = .
We correlate [ ]x n with the sin and cos curves of the frequency 2 kHz . These sin and cos
curves in Figure 5.5 have 2k = whole periods in the sampling interval (in previous case only
one).
0 1 2 3 4 5 6-1.5
-1
-0.5
0
0.5
1
1.5
k=2
Figure 5.5 Signal x(t) and sin, cos curves with 2 kHz used for sampling
Substituting the values of samples [ ]x n into the equation (5.35) for 2k = gives the result:
( ) 452 1,414 j l, 414 2 jX e= + =o
(5.39)
The input signal [ ]x n contains a term of frequency 2 kHz, its relative amplitude is 2 and its
phase angle to the cos curve of frequency 2 kHz is 45°.
In a similar manner, we compute the third component for 3k = .
For 3k = , we correlate [ ]x n with sin and cos line of the frequency 3 kHz (Figure 5.6).
54
0 1 2 3 4 5 6-1.5
-1
-0.5
0
0.5
1
1.5
k=3
Figure 5.6 Signal x(t) and sin, cos curves with 3 kHz used for sampling
We should notice again, that they have 3k = full periods in the sampling window.
Substituting the values of the samples x(n) to the equation (5.35) and for 3k = returns
( )3 0 – 0 0X j= = (5.40)
DFT yields that [ ]x n do not contain a term with the frequency 3 kHz.
We continue in a similar manner the computation of DFT for the subsequent values of the
parameter k .
For 4k = the result of computation according to (5.35) is
( )4 0,0— 0,0 0X j= = (5.41)
For 5k = the result of computation according to (5.35) is also zero
( )X 5 0,0— j0,0 0= = (5.42)
and curves used for 5k = are shown in Figure 5.7
0 1 2 3 4 5 6-1.5
-1
-0.5
0
0.5
1
1.5
k=5
Figure 5.7 Signal x(t) and sin, cos curves with 5 kHz used for sampling
55
For 6k = the result of computation according to (5.35) is non zero
( ) 452 1,414 j l, 414 2 jX e= + =o
(5.43)
and curves used for 6k = are shown in Figure 5.8.
0 1 2 3 4 5 6-1.5
-1
-0.5
0
0.5
1
1.5
k=6
Figure 5.8 Signal x(t) and sin, cos curves with 6 kHz used for sampling
Also for 7k = the result of computation according to (5.35) is non zero
( ) 907 0 4 4 jX j e= + =o
(5.44)
and curves used for 7k = are shown in Figure 5.9.
0 1 2 3 4 5 6-1.5
-1
-0.5
0
0.5
1
1.5
k=7
Figure 5.9 Signal x(t) and sin, cos curves with 7 kHz used for sampling
Finally, for 0k = we use for computation ( ) ( )cos 0 sin 0j− and obtain
( ) ( ) ( ) ( ) ( )1 1
0 0
0 cos 0 sin 0( (
n n
X x n j X n− −
= =
= − = ∑ ∑ (5.45)
substituting sampled data
56
( )0 0,3535 0,3535 0,6464 1,0607 0,3535 1,0607 1,3535 0,3535 0X = + + + + − − − = (5.46)
The value ( )0X in the frequency domain determines the constant term in the spectrum of
the signal [ ]x n .
Plotting the output values ( )X k as a function of frequency, we obtain an amplitude
spectrum of the input sequence [ ]x n (Figure 5.10).
We should notice, that the picture shows for the sequence [ ]x n , two spectral components 1
kHz (k = 1) and 2 kHz (k = 2). Additionally, the first component of the frequency 1 kHz has
twice as high amplitude as the second component with the frequency 2 kHz.
0 1 2 3 4 5 6 70
0.5
1
1.5
2
2.5
3
3.5
4
0 1 2 3 4 5 6 7-0.2
0
0.2
0.4
0.6
0.8
1
1.2
1.4
1.6
0 1 2 3 4 5 6 7-4
-3
-2
-1
0
1
2
3
4
0 1 2 3 4 5 6 7-100
-80
-60
-40
-20
0
20
40
60
80
100
Absolute
ValuePhase
RealIM
Img
Figure 5.10 DFT Results: abs value, phase, real and imaginary part of all eight components
Looking at Figure 5.10 two natural question may be asked.
Firstly, what is the meaning of amplitudes for k 6= and k 7= ? Secondly, why do the
amplitudes seem to be four times as high as we would expect them to be?
57
The answer to the two questions reveals two important features of the DFT. Every single
output value ( )X k is a sum of subsequent products of input sequence samples and the sin
and cos components, whose frequencies are so chosen, that they have k full periods in the
whole range of N samples. Additionally, the output terms of DFT are symmetrical.
5.4. Properties of DFT
We firstly take a closer look at the symmetry property. If the input sequence [ ]x n is real
then, the complex output values of DFT for arguments 2≥ (k are superfluous in regard to
output values for the arguments ranging from k 0= to N
k 12
= − . k-th Output value of DFT
has the same amplitude as the (N—k)-th output value. The phase angle of the m-th output
value of DFT is equal to the phase angle of (N—m)-th output value of DFT with the minus
sign.
We may compactly write the properties using complex conjugate numbers denoted with (*).
( ) ( )∗= −X k X ( k (5.47)
Let us once again take a look at the signal analysis presented above.
In the example under consideration, we see in Figure 5.10 that X(5), X(6) and X(7) are complex
conjugated values with X(3), X(2) and X(1), respectively. The real part of X(k) has the
property of even symmetry (see Figure 5.10) and imaginary part of DFT has the property of
odd symmetry (see Figure 5.10).
Therefore, to obtain the DFT of the signal [ ]x n , it is enough to compute the first N
2 values
of ( )X k , where 20 1≤ ≤ −(k . • If the real input function is even x(n)=x(-n), then X(k) is
always real and even. On the other hand, If the real input function is odd, -x(n)=x(-n), then
XRe(k) is always equal to zero, and XIm(k) is non zero.
Let us now take a closer look at the linearity property of DFT. It is a very important property,
because It says that the DFT of a sum of two components is equal to the sum of transforms
of each term independently. Substituting
( ) ( ) ( )1 2= +sumx n x n x n (5.48)
into the equation (5.1) in order to obtain ( )sumX k we get
58
( ) [ ] [ ]( )
[ ] [ ]
( ) ( )
12 /
1 2
0
1 12 / 2 /
1 2
0 0
1 2
−−
=
− −− −
= =
= + =
= + =
= +
∑
∑ ∑
(j kn (
sum
n
( (j kn ( j kn (
n n
X k x n x n e
x n e x n e
X k X k
π
π π (5.49)
The proper interpretation of the amplitudes of DFT requires more explanations. Once again,
we will consider the computational example presented in Figure 5.10.
The results |X(1)|=4, |X(2)|=2 are ambiguous, because the components of the input signal
[ ]x n had the amplitudes 1 and 0,5 respectively.
If the real input signal contains a sin component with the amplitude A0 and an integer
number of periods in the interval of N input samples then the amplitude of DFT for this
particular sin signal is Mr where
02
=r
(M A (5.50)
If the input signal for DFT contains a constant term D0, then the amplitude of the constant
output component X(0) DFT is equal to
( ) 0DX 0N
= (5.51)
Considering the previous case of the real input signal for the component with frequency
1000 Hz A0=1 and N=8, we obtain
r
8M 1 4
2= ⋅ = (5.52)
as shown in our example.
In the literature on DFT, the transformation is sometimes defined with an included scaling
term
( ) [ ]1
2 /
0
2 −−
=
= ∑(
j nk (
n
X k x n e(
π (5.53)
The scaling term 2/N guaranties, that the amplitudes of X(k) are equal to the amplitudes of
the sin components in the input signal in time domain. Unfortunately, there is additional
computational “cost” included in the operation – division by N.
Commercial software packages include often the following relationships:
( ) [ ]1
2 /
0
1 −−
=
= ∑(
j nk (
n
X k x n e(
π (5.54)
59
and
[ ] ( )1
2 /
2 2
0
1 −
=
= ∑(
j nk (
k
x n X k e(
π (5.55)
for direct and inverse DFT.
The scaling terms 1
N seem a little bit ambiguous, but they are in use, to avoid the change
of scaling coefficients by transforming in both directions.
5.5. Shifting theorem in DFT
There is another property of the DFT, known as shifting theorem. As it has tremendous
impact on practical computations a whole chapter has been dedicated to it. The shifting
theorem states that:
The shift in time of a periodic input sequence x(n)
manifests itself as a constant phase shift in the angles associated with DFT results.
If we decide to sample [ ]x n beginning with n equal to a particular constant n = k, as a
opposition to n = 0, then DFT of the time shifted samples is given by
( ) ( )2 /= j km (
kX m e X mπ (5.56)
From (5.56) we see that if the point at which we start sampling [ ]x n is shifted to the right
by k samples, then the output spectrum Xk(m) of DFT is expressed X(m), with every complex
term X(m) multiplied by a linear phase shift 2 /j km (e π
, actually shifting the phase by
2 km / Nπ .
On the other hand, if the point at which we start to sample x(n) is shifted to the left side by k
samples, then the spectrum Xk(m) is expressed as X(m) multiplied by 2 /− j km (e π
.
In order to illustrate the shifting phenomenon, the same discrete signal from the previous
example has been sampled with a delay (or more precisely a shift) of k 3= samples.
( ) ( ) ( )312 4
sin 2 1000 sin 2 2000= ⋅ ⋅ + ⋅ ⋅ +x t t t ππ π (5.57)
Figure 5.11 shows the original signal defined through (5.57) in the time domain.
60
0 0.2 0.4 0.6 0.8 1 1.2 1.4
x 10-3
-1.5
-1
-0.5
0
0.5
1
1.5
0
1
2
3
4
5 6
7
Figure 5.11 Sampling of x(t) in both examples
The new, shifted series x(n) contains points labeled with numbers (see picture) with the
values:
x(0) =1,0607, x(1) =0,3535 ,
x(2) = - 1,0607, x(3) = - 1,3535,
x(4) = - 0,3535, x(5) = 0,3535,
x(6) = 0,3535, x(7) = 0,6464
Computing DFT of the above series, Xk(m) we obtain results given in Table 5-1 and presented
graphically in
61
Figure 5.12.
Table 5-1 Shifted DFT values
m amplitude phase real part imaginary part
0 0 0 0 0
1 4 +45 2,8284 2,8284
2 2 -45 1,4141 -1,4141
3 0 0 0 0
4 0 0 0 0
5 0 0 0 0
6 2 +45 1,4141 -1,4141
7 4 -45 2,8284 -2,8284
62
0 2 4 6 80
1
2
3
4
0 2 4 6 8-4
-2
0
2
4
0 2 4 6 8-1
0
1
2
3
0 2 4 6 8-4
-2
0
2
4
ABS PHASE
REAL PART IMAGINARY PART
Figure 5.12 DFT Results for the 2 example: absolute value of Xk(m), phase Xk(m), real part Xk(m), imaginary part
Xk(m).
From the computation results we see, that the amplitude of Xk(m) is unchanged in relation
to the amplitude of X(m) of the non-shifted signal. The observations may be presented
briefly:
The amplitude of DFT related to the original signal did not change despite sampling in shifted
interval.
The phase of DFT, however, changes with relation to the moment when we start sampling of
the signal x(n).
Now, considering the component of DFT Xk(m), for m = 1 we check the phase of the shifted
signal.
Recalling that X(1) from the 1 example DFT had an amplitude 4 and the phase -90 we have
for k = 3 and N = 8:
( ) ( )3
2 28 2 4
3 1 1 4 4−
= ⋅ = ⋅ =m
j k j j j(X e X e e e
π ππ π
(5.58)
Therefore, X3(m) has an amplitude 4 and phase +45 degree.
63
5.6. Inverse Discrete Fourier Transform
There is a simple way to come back from the spectral components to the original signal in
time domain. The standard expression for inverse discrete Fourier Transform IDFT is given
shortly in the exponential notation
( ) ( )1
2 /
0
1 −
=
= ∑(
j mn (
m
x n X m e(
π (5.59)
or applying the sin and cos terms
( ) ( )1
0
1cos 2 sin 2
−
=
= +
∑(
m
n nx n X m m j m
( ( (π π (5.60)
Discrete time signals may be regarded as a sum of sinusoidal terms with various values of
frequencies. The values X(m) of DFT are N complex values, determining the amplitude and
phase of every term of the sum.
The inverse transform will be explained further on an example.
If we compute IDFT substituting the results from the 1 example into the equation (3), then
we will go back from the frequency domain to the time domain and we will obtain the values
of the original signal x(n).
[ ] ( )
[ ] ( ) ( ) ( ) ( ) ( )( )[ ]
12 2/8
0
0/8 4/8 6/8 8/8 14/8
12
8
12 0 1 2 3 ... 7
8
2 0,6464
−
=
=
= + + + +
=
∑(
j m
m
j j j j j
x X m e
x X e X e X e X e X e
x
π
π π π π π (5.61)
and
x[0] = 0,3535, x[1] = 0,3535, x[2] = 0,6464, x[3] = 1,0607,
x[4] = 0,3535, x[5] = -1,0607, x[6] = -1,3535, x[7] = -0,3535.
We should notice that the expression for IDFT, given by the equation (3), differs from the
equation for DFT only with the scaling term 1/N and the change of the sign of the exponent.
Apart from the difference in scaling of values, all properties of DFT, may also be related to
the IDFT.
5.7. DFT Leakage (Ambiguity) and its minimization
The previous examples of DFT have shown accurate results, because input sequences x(n)
had been chosen carefully from a set of sinusoidal signals.
64
In a general case, DFT of sampled signals may lead to ambiguous results in the frequency
domain.
The DFT property, known as spectrum leakage, results in the fact that DFT values are only
an approximation of the spectrum for signals after the sampling (discretization).
There are methods minimizing that effect, but it is not possible to eliminate it completely.
DFT is bounded to operate on a finite number N of input samples, which are sampled with
the frequency fp. As a result we get N- point transform for discrete values of frequency:
; 0,1,2,..., 1= = −p
m
mff m (
( (5.62)
for which the values of DFT are computed.
DFT gives accurate results, only if the input sequence energy is located exactly at
frequencies given in equation (5), for which we do the analysis. That are integer multiplies
of the fundamental frequency fp/N.
Presented theoretical dependencies will be illustrated with a practical computational
example.
If the input signal contains a term with an intermediate frequency e.g. 1,5 fp / N between
the values mfp/N for which we compute DFT, then the input component will be visible by all
N output values of DFT.
Let’s compute a 64-points DFT for a series obtained after sampling three periods of a
sinusoid (Figure 5.13). Computed DFT shows that the signal do not contain a component
with a frequency other then m=3. The correlation of an input sequence and sinusoids for m
different from 3 is equal to zero.
0 2 4 6 8 10 12 14 16 18 20 22-1.5
-1
-0.5
0
0.5
1
1.5
0 5 10 15 20 25 30 350
5
10
15
20
25
30
35
3 periods DFT
65
Figure 5.13 64-points DFT (left) input sequence, (right) absolute value of DFT, the first part of
the result
We have now an input series containing 3,4 periods for 64 samples. Because this input
sequence does not have an integer number of periods represented by the 64 samples, the
energy “leaks” to all commuted samples of DFT, as shown in Figure 5.14 (right).
0 2 4 6 8 10 12 14 16 18 20 22-1.5
-1
-0.5
0
0.5
1
1.5
0 5 10 15 20 25 30 350
5
10
15
20
25
30
3,4 Periods DFT
Figure 5.14 64-points DFT (left) input sequence, (right) absolute value of DFT the first part of
the result
DFT sample e.g. for m = 4 is not equal to zero, because, the sum of products of the input
series and the term for frequency analysis at m= 4 is not equal to zero.
That is a leakage, which means every input signal with a frequency not equal to the
particular frequency for which the DFT has been computed at the moment, leaks to all other
DFT samples.
The leakage effect is inevitable when we compute DFT of a real time series with a finite
length.
Knowing the reason and importance of leakage we should consider the minimization
possibilities. Try to answer the question how to predict and minimize the leakage effect.
For a real sinusoidal signal with k periods in N- points input time sequence, the values of
frequency stripes of N-points DFT as a function of index m are approximated with the
function sinc (sinus cardinalis)
( )( )( )
sin
2
− = ⋅−
k m(X m
k m
π
π (5.63)
66
-5 -4 -3 -2 -1 0 1 2 3 4 5-0.4
-0.2
0
0.2
0.4
0.6
0.8
1
Widmo ciągłe
( )( )
( )sin
2
k m(X m
k m
π
π
−= ⋅
−
DFT
coninuous spectrum
-5 -4 -3 -2 -1 0 1 2 3 4 50
0.5
1
( )( )
( )sin
2
k m(X m
k m
π
π
−= ⋅
−
DFT
Widmo amplitudoweAmplitudespectrum
Figure 5.15 DFT for N-point input sequence, containing k periods of real sinusoid: (above)
spectrum as a function of m-th sample DFT, (below) amplitude spectrum.
67
The curve in the picture 5 (above), containing the main lobe and periodic picks and walleyes
known as side lobes, may be regarded as N point spectrum of real sinusoidal time series
with k full periods in N-points input time interval.
The DFT values are discrete samples, which are on the curves in Figure 5.15, which means
DFT will be the sampled version of the continuous spectrum.
If the input signal has exactly an integer number of periods k, the leakage will not appear.
The reason for that: the angle in the numerator of equation (5.63) is an integer multiply of
π , then the sinus of it is equal to zero.
If the input sinusoid has an integer number of periods in the interval of the length N, then
the output values of DFT are located on the continuous spectrum curve exactly on the zero
crossing points of that curve.
As usual, an example will explain in detail the presented theory.
A real sinusoid with the frequencies 8 kHz,8,5 kHz,8,75 kHz, and amplitude 1, has been
sampled with the frequency 32kHz. For 32-points DFT the distance between samples is
fp/N=1kHz. DFT results are shown in Figure 5.16 and Figure 5.17.
3000 4000 5000 6000 7000 8000 9000 10000 11000 12000 130000
2
4
6
8
10
12
14
16
18
20
8kHzDFT
Figure 5.16 DFT for 32-points sinusoidal input sequence , signal freguency f=8kHz,
68
3000 4000 5000 6000 7000 8000 9000 10000 11000 12000 130000
2
4
6
8
10
12
14
16
18
20
8,75kHzDFT
3000 4000 5000 6000 7000 8000 9000 10000 11000 12000 130000
2
4
6
8
10
12
14
16
18
20
8,5kHzDFT
Figure 5.17 DFT for 32-points sinusoidal input sequence (above) signal freguency f=8,5kHz, (below) f=8,75kHz
The results of DFT leakage are troublesome, because the values of stripes correlated with
low amplitude components will be distorted by side lobes of nearby stripes for signals with
high amplitude.
An important technique, known as windowing, is the most common way of reducing the
leakage. Windowing is reducing the leakage by minimizing the amplitude of side lobes of the
function sinc (equation (5.63)).
69
-20 0 20 40 60 80 100
-1
0
1
-20 0 20 40 60 80 100
-1
0
1
-20 0 20 40 60 80 100
-1
0
1
Figure 5.18 Minimizing discontinuity at the ends of sampling intarval: (above) input infinite sinusoid; (middle) a
square window adequate to the sampling interval, (bottom) the product of the square window and input
sinusoid.
-20 0 20 40 60 80 100
-1
0
1
-20 0 20 40 60 80 100
-1
0
1
-20 0 20 40 60 80 100
-1
0
1
Figure 5.19 (above) input sinusoid of infinite duration; (middle) triangular window function , (bottom)
multiplication of the signal and window
70
-20 0 20 40 60 80 100
-1
0
1
-20 0 20 40 60 80 100
-1
0
1
-20 0 20 40 60 80 100
-1
0
1
Figure 5.20 (above) input sinusoid of infinite duration (middle) Hanning window, (bottom)
multiplication of Hanning window and input sinusoid
Let’s consider a signal of infinite duration in time domain shown in Figure 5.18 (above). DFT
may be done only on a finite time sequence, as shown in Figure 5.18 (bottom). We may
regard DFT of the input signal in Figure 5.18 (bottom) as DFT of the product of infinite input
signal form Figure 5.18 (above) and a square window with the amplitude 1 on Figure 5.18
(middle).
Every time, when computing DFT of a finite time series (bottom in Figure), by default we
multiply the sequence in sampling interval by a window of ones and other values by zeros.
As it turns out, the shape of the function sinc=sin(x)/x is caused by the square window,
because the DFT of a square window is a sinc function.
To minimize the leakage effect caused by the side lobes we must minimize their
amplitudes by applying other windows then a square one.
Imagine that we have multiplied the input signal form Figure 5.19 (above) by a triangular
window from Figure 5.19 (middle), in order to obtain a windowed signal shown in picture
Figure 5.19 (bottom).
When you look at the Figure 5.19, you notice that the values of the final input signal are
equal on both sides of the sampling interval. Reduced discontinuity decreases the level of
71
relatively high frequency components in the whole DFT values set. That means, the values of
side lobes have reduced amplitude due to the application of a triangle window.
There are other window types, which reduce the leakage even better, then the triangle
window such as Hanning window form Figure 5.20 (middle). Multiplication of the window in
Figure 5.20 (middle) by the input sequence gives us the signal shown in Figure 5.20 (bottom),
being the final input for DFT.
5.8. Windows types and their properties
To minimize the leakage effect different types of windows are in use.
Assuming, that N original samples of the input signal are indexed by n, where 0 1n (≤ ≤ − ,
we assign N coefficients of the window as w(n), which means, the input sequence x(n) is
multiplied by the window w(n), before we start DFT. Therefore DFT Xw(m) of the windowed
input signal x(n) has the form:
( ) ( ) ( )1 2
0
−
=
= ⋅∑n( j m(
W
n
X m w n x n eπ
(5.64)
- square window also called boxcar
( ) 1, for 0,1,2,..., 1= = −w n n ( (5.65)
- triangle window (similar to Barletta and Parzen windows)
( ); 0,1,2,..., / 2
/ 2
2 ; / 2 1, / 2 2,..., 1/ 2
== − = + + −
nn (
(w n
nn ( ( (
(
(5.66)
- Hanning window (called also lifted cosinus, Hanna or von Hanna)
( ) 1 1cos 2 , 0,1,2,..., 1
2 2
= − = −
nw n n (
(π (5.67)
- Hamming window
( ) 0,54 0, 46cos 2 ; 0,1,2,..., 1 = − = −
nw n n (
(π (5.68)
The spectrum of a square window is a measure to evaluate and compare different types of
windows.
There is a definition for a logarithmic amplitude response which helps to normalize and
compare spectra of various windows.
72
The width of main lobes of various windows depredates the spectral resolution of windowed
DFT almost twice. However, the advantages of suppressing leakage overweight the
worsening of spectral resolution
0 50 100 150 200 250 30010
-4
10-3
10-2
10-1
100
square
triangular
Hamming
Hanning
Figure 5.21 Absolute values of window responses in normalized logarithmic scale
We notice a reduction of first side lobe and drastic reduction of further lobes in Hanning
window. Hamming window has even a smaller value of the fist side lobe, but the further
lobes do not increase as efficiently as Hanning.
That means, the leakage three or four stripes away from main stripe is smaller for Hamming
window then for Hanning. But the leakage further away shows an opposite dependence.
If we apply the Hanning window to the example of 3.4 periods in sampling interval, we
obtain DFT output values for this windowed signal along with the results for square window.
See
73
Figure 5.22.
Figure 5.22 Comparison of DFT results for square and Hanning windows
As awaited, the amplitude spectrum for Hanning window is brighter and has a lower
maximal value, but the leakage of side lobes is significantly smaller in comparison to the
leakage of square window.
Therefore, we may say that the choice of a window is a compromise between the
brightening of the main lobe, value of the first lobe and the decrease rate of side lobes with
increasing frequency. The application of a certain window depends on a particular situation.
5.9. Resolution of the DFT, filling with zeros, sampling in the frequency
domain
One of the popular methods of improving frequency resolution of DFT is a method called
zero filling.
74
This process requires an addition of zero samples to the original input sequence in order to
increase the number of input samples.
Sampling a function continuous in time domain, having a continuous Fourier transform, and
afterwards computing DFT from these samples, we see that this DFT gives us sampled
approximation of the continuous Fourier transform in frequency domain.
We may assume that the more samples in the DFT, the better the approximation of the
continuous transform is.
Figure 5.23 Sampling of DFT in frequency domain: 16 samples of input data and N = 16;
Figure 5.24 Sampling of DFT in frequency domain: 16 samples of input data, 3x16 added zeros
75
0 20 40 60 80 100 120 140 160 180-1
-0.8
-0.6
-0.4
-0.2
0
0.2
0.4
0.6
0.8
1
10x16 zerowych próbek
0 10 20 30 40 50 60 70 80 900
1
2
3
4
5
6
7
8
Próbki wejściowe Moduł DFT
Figure 5.25 Sampling of DFT in frequency domain: 16 samples of input data, 10x16 added zeros
Adding zeros to the input sequence improves the spectral resolution of DFT, but there is a
practical border determining how much we can gain through adding zero samples.
In practice, if we want to introduce both zero filling and windowing, we should be careful
not to apply a window to the whole signal after addition of zeros.
5.10. Literature
[1] Angelo Baggini, Handbook of power quality, Wiley, Chichester, 2008, pp. 187-259
[2] R. G. Lyons Understanding digital signal processing Addison Wesley Longman, 1997
[3] S.J. Orfandis, Introduction to Signal Processing, Prentce Hall, 1996
[4] A. V. Oppenheim, R. W. Schafer, Discrete-Time Signal Processing, Prentce Hall, 1999
76
6. Symmetrical components for current and voltage unbalance
assessment
Three phase voltages and currents should be symmetrical during the operation of a
power system. Therefore, the voltages at generators taps are of the same magnitude and
displaced equally to each other at an angle of 120 degrees. Components of the power
system (lines, transformers, switch gears, etc. ) are designed and build in order to guarantee
linearity and symmetry. The greatest difficulty brings the appropriate connection of single
phase loads differing significantly in operational characteristics. Attention must be paid to
equal distribution of them.
In everyday practice it is hardly possible to obtain an ideal voltage symmetry at all nodes
of the electrical sources. Similarly, currents are very often unsymmetrical.
The main sources of unbalance in voltage and current are low-voltage and medium
voltage one phase appliances. The operation of single-phase loads results in various currents
in all feeding lines. As a consequence, unequal voltage drops appear on system nodes. An arc
furnace is an example of a three-phase load introducing unbalance conditions. The
characteristic of a melting process evolves different impedances of high current paths.
System components, especially overhead lines contribute to the unbalance
phenomenon. The geometry of most popular electrical masts forces an unequal position of
the conductors in relation to each other and to the earth.
Faults, excluding three phase balanced faults, and other abnormal operation
condition contribute additionally to the levels of unbalance.
In this chapter the mathematical description of the unbalance phenomena will be
presented. Unbalance factors and border values given in standards will be presented.
The theoretical ideas of the chapter are from [7].
77
6.1. Mathematical Background
The concluding idea derived from a three phase system analysis suggests that it is
much easier to deal with a symmetrical system than with a nonsymmetrical one.
Computation of currents, voltages and powers is conceptually less demanding and more
stride forward.
6.1.1. Complex transformation parameter a
Generally, the analysis of an unsymmetrical electrical system may be transformed
into an analytical problem of a special set of symmetrical systems. The superposition of
results gives the original currents and voltages.
In other words, the symmetrical component method is basically a modelling
technique that permits a systematic analysis and design of three phase systems. Decoupling
a detailed three phase network into three simpler sequence networks reveals complicated
phenomena in more simplistic terms. Sequence network results can then be superimposed
to obtain three phase results. See Figure 6.1.
Figure 6.1 Positive sequence, negative sequence and zero sequence voltage systems
To express quantities algebraically, the complex operator a is used. This denotes a
phase shift by +120 degree and multiplication by unit magnitude.
120 3cos120 sin120 0.5
2
ja e j j°= = °+ ° = − + (6.1)
therefore
2
3
20.5
3
1
a j
a
= − −
=
(6.2)
Geometric addition of three vectors 2 3 a a a gives us an equilateral triangle
21 0a a+ + = (6.3)
78
Basic algebraic dependences for the a operator are given in expressions
2 3 30,5 0.5 3
2 2a a j j j
− = − + − − − =
(6.4)
and
21 3a ja− = (6.5)
21 3a ja− = − (6.6)
Equations (6.4) to (6.6) have an geometric representation on the complex numbers plane.
Figure 6.2 geometrical representation of algebraic operations on vectors 2 a a
6.1.2. Symmetrical components definition
Applying the a operator of the three voltages may be resolved into three sets of
sequence components; according to
2
shift kn
π= − (6.7)
where n is the number of phases and k=0,1,2. (Figure 6.1)
• Positive Sequence
The voltage in the subsequent phase has a negative displacement of 120 degree
(213
shiftπ
= − )
( )( )( )
1 1
2 1
3 1
sin
sin 120
sin 240
L
L
L
u U t
u U t
u U t
ω
ω
ω
°
°
=
= −
= −
(6.8)
79
1 1
2
2 1
3 1
L
L
L
U U
U a U
U aU
=
= =
(6.9)
• Negative Sequence
The voltage in the subsequent phase has a negative displacement of 240 degree
(2
23
shiftπ
= − )
( )( )( )
1 2
2 2
3 2
sin
sin 240
sin 120
L
L
L
u U t
u U t
u U t
ω
ω
ω
°
°
=
= −
= −
(6.10)
1 2
2 2
2
3 2
L
L
L
U U
U aU
U a U
=
= =
(6.11)
• Zero Sequence
The voltage consists of three phasors with equal magnitudes and with zero phase
displacement between them (233
shiftπ
= − ).
( )( )( )
1 0
2 0
3 0
sin
sin
sin
L
L
L
u U t
u U t
u U t
ω
ω
ω
=
= =
(6.12)
1 0
2 0
3 0
L
L
L
U U
U U
U U
=
= =
(6.13)
The load (Figure 6.3) is powered in the same manner, regardless of the switch position. A
cascade connection of three symmetrical generators (zero, positive and negative sequence)
constitutes an intuitive equivalent of an unsymmetrical generator G.
80
Figure 6.3 Representation of an unsymmetrical generator G with a cascade connection of zero sequence
generatorG0, positive sequence generator G1 and negative sequence generator G2
6.1.3. Symmetrical components of voltage, current, impedance and
admittance
Mathematically, the idea presented in Figure 6.3 is expressed by a set of equations
1 0 1 2
2
2 0 1 2
2
3 0 1 2
L
L
L
U U U U
U U a U aU
U U aU a U
= + +
= + + = + +
(6.14)
or in matrix notation
01
2
2 1
2
3 2
1 1 1
1
1
L
L
L
UU
U a a U
a aU U
=
(6.15)
S - matrix is called a symmetrical components transformation matrix
2
2
1 1 1
1
1
a a
a a
=
S (6.16)
The determinant of the S matrix
( )2det 3 3 3a a j= − =S (6.17)
is non zero, thus S is a nonsingular matrix (it is invertible).
Computing the reverse transformation we obtain
0 1
2
1 2
2
2 3
1 1 111
31
L
L
L
U U
U a a U
a aU U
=
(6.18)
writing down (6.18) similarly as (6.14) we get
81
( )
( )
( )
0 1 2 3
2
1 1 2 3
2
2 1 2 3
1
3
1
3
1
3
L L L
L L L
L L L
U U U U
U U a U aU
U U aU a U
= + +
= + +
= + +
(6.19)
The zero sequence, positive sequence and negative sequence components are obtained
directly from phase voltages and contain exactly the same information.
Transformation procedure may be also applied for three phase currents
01
2
2 1
2
3 2
1 1 1
1
1
L
L
L
II
I a a I
a aI I
=
(6.20)
and reverse transformation
0 1
2
1 2
2
2 3
1 1 111
31
L
L
L
I I
I a a I
a aI I
=
(6.21)
Transformation of currents and voltages are compactly given in matrix notation
1 1
1 1 U
3 3s s s s= = = =U SU S U I SI I S I (6.22)
where
1 3= ⋅SS I (6.23)
for impedances and admittances we have similar matrix notation
1 1
1 1
3 3s s s s= = = =Z Sz z S Z Y Sy y S Y (6.24)
6.1.4. Basic properties of symmetrical components
• Three phase source or load connected in wye and without a fourth zero line indicates
no zero sequence current component 0 0I = . It is a consequence of Kirchhoff law for
the node of wye connection ( )0 1 2 3
1
3L L LI I I I= + + .
• In a three phase system of voltages and currents where only one voltage or current is
different from zero the three symmetrical components are equal and amount exactly
82
to 1
3 of the line voltage or current. For 1 0LI ≠ and 2 30, 0L LI I= = we get
0 1 2 1
1
3LI I I I= = = .
• In a three phase system of voltages and currents where only one voltage or current is
exactly zero the geometrical sum of three symmetrical components is zero
0 1 2 0I I I+ + = .
• In a three phase system of voltages with or without neutral the zero sequence
component of L-L voltage is zero 0 0U = . It is a consequence of a fact that the
geometrical sum of three L-L voltages is zero.
• In the three phase delta connected source or load windings can circulate the zero
sequence current component. It is not present in the line currents.
6.2. Ohm’s low for symmetrical components
A three phase unsymmetrical load is given in Figure 6.4. Starting from the three equations
defining the relation between voltage and current (6.25) we formulate the Ohm’s low in a
matrix form
Figure 6.4 Three phase unsymmetrical load
A A A
B B B
C C C
U Z I
U Z I
U Z I
=
= =
(6.25)
0 0
0 0
0 0
A AA
B B B
CC C
U IZ
U Z I
ZU I
= =
(6.26)
where 'Z is the diagonal impedance matrix
83
0 0
' 0 0
0 0
A
B
C
Z
Z
Z
=
Z (6.27)
using the matrix notation
'=U Z I (6.28)
Substituting s s= =U SU I SI and multiplying both sides of (6.28) by 1−S we obtain
1 's s
−=U S Z SI (6.29)
The Ohm’s law for symmetrical components
's s=U z I (6.30)
where
1' 's
−=z S Z S (6.31)
after evaluation of (6.31) we get
0 2 1
1 0 2
2 1 0
'
z z z
z z z
z z z
=
z (6.32)
Finally, a formula (6.25) corresponds with a set of scalar equations for symmetrical
components
0 0 0 2 1 1 2
1 1 0 0 1 2 2
2 2 0 1 1 0 2
U z I z I z I
U z I z I z I
U z I z I z I
= + +
= + + = + +
(6.33)
Computation of every symmetrical component of voltage requires all symmetrical
components of the current.
Similar concepts may be developed for admittances, therefore we rewrite (6.25) introducing
admittances
A A A
B B B
C C C
I Y U
I Y U
I Y U
=
= =
(6.34)
and apply the same procedure described above, finally obtaining
0 0 0 2 1 1 2
1 1 0 0 1 2 2
2 2 0 1 1 0 2
I y U y U yU
I yU y U y U
I y U yU y U
= + +
= + + = + +
(6.35)
84
Computation of every symmetrical component of current requires all symmetrical
components of the voltage.
Generally, the symmetrical component admittance is not an inverse of impedance and is
given by
2
0 1 20 3 3 3
0 1 2 0 1 2
2
2 0 11 3 3 3
0 1 2 0 1 2
2
1 2 02 3 3 3
0 1 2 0 1 2
3
3
3
z z zy
z z z z z z
z z zy
z z z z z z
z z zy
z z z z z z
−= + + −
−=
+ + − − =
+ + −
(6.36)
6.3. Kirchhoff’s current low and voltage low for symmetrical components
A three phase wye – wye connected system is shown in Figure 6.5. Kirchhoff’s current low is
a linear relationship between currents. The sum of currents flowing into N1 node is equal to
the sum of currents flowing out of that node. The transformation of current into symmetrical
components (6.21) is also a linear operation. Therefore, the same relationship is valid for
currents transformed (or not) into symmetrical components. No further considerations are
needed.
Figure 6.5 A three phase wye - wye connected system.
Kirchhoff’s voltage low requires more attention. We start with a selection of three mashes,
each consisting from the phase line and neutral. The voltage in every closed loop must be
equal to zero, therefore
A A ( B B ( C C (E U U E U U E U U= + = + = + (6.37)
85
The value of source voltage is equal to a voltage drop on phase impedance and the neutral.
The first term will be rewritten with accordance to (6.25), corresponding symmetrical
components are given by (6.33). The voltage drop on the neutral can be rewritten into
( ) 03( ( A B C (U Z I I I Z I= + + = (6.38)
Equation (6.38) includes only one symmetrical component.
Finally, the Kirchhoff’s voltage low for symmetrical components is given by
0 00 2 1
1 1 0 2 1
2 1 02 2
3 (E Iz Z z z
E z z z I
z z zE I
+ =
(6.39)
in compact form
,,
s s s=E z I (6.40)
where
0 2 1
,,
1 0 2
2 1 0
3 (
s
z Z z z
z z z
z z z
+ =
z (6.41)
Comparing the square matrices ,
sz and ,,
sz in the notations of Ohm’s law and Kirchhoff’s law
we see that there is only one different entry at (1,1). Therefore, there is apparent equality
between the positive and negative symmetrical components of the source and load voltages
1 1 2 2 E U E U= = .
Especially for a symmetrical load ( A B CZ Z Z= = ) we observe a significant simplification. The
positive and negative symmetrical component of impedances is zero and we have only
0 Az Z= . The matrix equation (6.39) can be rewritten to
( )0 0 0
1 0 1
2 0 2
3 (E z Z I
E z I
E z I
= +
= =
(6.42)
6.4. Filtering of symmetrical components
Filters of symmetrical components are used in order to deliver adequate values for further
processing. Basic ideas and concepts of filtering are presented below.
86
6.4.1. Zero sequence filtering
In a three phase system the neutral wire s conducting a current three times higher than the
zero component. Therefore, an ammeter indicates 03I (Figure 6.6). There is no need for
further filtering. However, in a network without neutral wire three current transformers are
needed (Figure 6.6). The ammeter shows 03kI .
Figure 6.6 Zero sequence component filters in networks with neutral (left) and without neutral (right)
Examples of a voltage symmetrical component filter are shown in Figure 6.7.
Figure 6.7 voltage symmetrical components filters
In order to filter zero sequence component a wye – connected identical impedances Z are
needed (Figure 6.7 left). The artificial neutral point 1N is connected with neutral wire via NZ
impedance. According to the symbols in Figure 6.7 we get
A B C (I I I I+ + = (6.43)
the sum of voltages may be therefore expressed as
( )A B C A B C (V V V Z I I I ZI+ + = + + = (6.44)
rewriting formula (6.44) we obtain an expression for the voltage across neutral wire
( )(( ( ( A B C
ZU Z I V V V
Z= = + + (6.45)
The line voltages
87
A A (
B B (
C C (
U V U
U V U
U V U
= +
= +
= +
(6.46)
after summation yield
3A B C A B C (U U U V V V U+ + = + + + (6.47)
so, the zero component is given by
( ) ( )0
1 1
3 3A B C A B C (U U U U V V V U= + + = + + + (6.48)
Comparing (6.44) and (6.48) we can express (U voltage as a function of the zero sequence
component
0
3
3
((
(
ZU U
Z Z=
+ (6.49)
and the current in the neutral wire
0
3
3
((
( (
UI U
Z Z Z= =
+ (6.50)
is proportional to the zero sequence component in the voltage.
Figure 6.7 (right) shows an example of a zero sequence filter where the voltage is measured
on the secondary side of a wye–delta measuring transformer. The voltages on the secondary
side are , ,AB BC CAU U U therefore the zero sequence component is given by
( )0
1
3AB BC CAU U U U= + + (6.51)
6.4.2. Positive and negative sequence components filtering
There are many approaches to the filtering of positive and negative sequence components.
In order to illustrate the question, two filters are presented as examples.
Figure 6.8 shows a three phase line without neutral wire feeding a wye connected load. Two
equal measuring transformers are installed in phase A and C. Transformed currents are
denoted as ,A CI I . Parallel to the current transformers two impedances are connected
1 2,Z Z , whose values will be determined further on. The secondary circuit is closed via
fZ impedance representing an ammeter or a transducer leading fI current. A brief analysis
of the circuit will be presented.
88
Figure 6.8 positive and negative sequence component filter for voltage and current
The sum of voltages in a mash is given by
( ) ( )1 2 0A f C f f fZ I I Z I I Z I− + − − = (6.52)
the current fI as a function of ,A CI I is given by
1 2
1 2
A Cf
f
Z I Z II
Z Z Z
+=
+ + (6.53)
In a three phase system without neutral the zero sequence is not present, therefore
1 2
21 2
A
C
I I I
I aI a I
= +
= + (6.54)
Then, we apply (6.54) to (6.53) and obtain an expression for filter current fI as a function of
the positive 1I and negative 2I sequence component
( ) ( )1 2 1 1 2 2
1 2f
f
Z aZ I Z aZ II
Z Z Z
+ + +=
+ + (6.55)
And now there are two approaches.
• Measuring positive sequence component in the current:
where the condition for correct measurement is
2
1 2 0Z a Z+ = (6.56)
and also
2
1 2Z a Z= − (6.57)
One of the solutions is 2Z R= than ( )601 1 3
2
j RZ e R j= = + . The resistance of the coil is
12
RR = and the reactance 1 3
2
RX = .
• Measuring negative sequence component in the current:
where the condition for correct measurement is
89
1 2 0Z aZ+ = (6.58)
and also
1 2Z aZ= − (6.59)
One of the solutions is 1Z R= than ( )602 1 3
2
j RZ e R j= = + . The resistance of the coil is
22
RR = and reactance 2 3
2
RX = .
Figure 6.9 shows a circuit for the measurement of positive and negative sequence
components in the voltage. The three phase system may be with or without the neutral wire.
Two identical measurement voltage transformers give voltages between phase A and B and
phase B and C, denoted with ,AB BCU U . Two impedances 1 2,Z Z , whose values will be
determined further on, are connected in series with the measurement transformers. The
two secondary mashes have a common branch denoted as fZ impedance representing an
ammeter or transducer leading fI current.
Figure 6.9 positive and negative sequence component filter of the voltage
The voltages in the two mashes in Figure 6.9 are given by
1
2
AB f f AB
BC f f BC
Z I Z I U
Z I Z I U
+ =
+ = (6.60)
The current fI is also a function of the line to line voltages
2 1
1 2 2 1
AB BCf
f f
Z U Z UI
Z Z Z Z Z Z
+=
+ + (6.61)
A symmetrical component of line to line voltages is equal to zero. Therefore
90
1 2
21 2
AB
BC
U U U
U a U aU
= +
= + (6.62)
substituting (6.62) to (6.51) we obtain an expression of the filter current fI as a function of
symmetrical components 1 2,U U of the line to line voltages.
2
2 1 2 11 2
1 2 2 1 1 2 2 1f
f f f f
Z a Z Z aZI U U
Z Z Z Z Z Z Z Z Z Z Z Z
+ += +
+ + + + (6.63)
And now there are two approaches.
• Measuring of a positive sequence component in the line to line voltage:
The condition for correct measurement is
2 1 0Z aZ+ = (6.64)
also
2
2 1Z a Z= − (6.65)
One of the solutions is 2Z R= than ( )601 1 3
2
j RZ e R j= = + . The resistance of the coil is
12
RR = and the reactance 1 3
2
RX = .
• Measuring of a negative sequence component in the line to line voltage:
The condition for correct measurement is
2
2 1 0Z a Z+ = (6.66)
and also
2
2 1Z a Z= − (6.67)
One of the solutions is 1Z R= than ( )602 1 3
2
j RZ e R j= = + . The resistance of the coil is
22
RR = and reactance 2 3
2
RX = .
Substituting the values proposed above into (6.63) we obtain simplified expressions for fI .
A positive sequence component
2
11
3
3f
f
jaI U
Z ja Z=
− (6.68)
A negative sequence component
91
221
3
3f
f
jaI U
Z ja Z=
+ (6.69)
6.5. Determination of unbalance with accordance to international
standards
Symmetrical components are primarily used for the analysis of different types of faults. Line
to line faults with and without earth connection, line to earth faults, symmetrical faults, etc.
Power quality assessment requires the implementation of symmetrical components for the
computation of asymmetry factor, mainly based on [3].
6.5.1. Unbalance factor
An unbalance factor “K” stands for the measure of unbalance. It is a ratio of negative and
(or) zero sequence component to the positive sequence component. The former is, the
negative sequence
22
1
100%U
UK
U= ⋅ (6.70)
22
1
100%I
IK
I= ⋅ (6.71)
and the latter is the expression for a factor related to zero sequence
00
1
100%U
UK
U= ⋅ (6.72)
00
1
100%I
IK
I= ⋅ (6.73)
Equations from (6.70) to (6.73) refer only to the main component of the voltage or current.
For isolated neutral power networks, the term “asymmetry” is more frequently used, it is
defined by the ratio of negative to positive sequence only.
The negative sequence component is transformed by a transformer irrespective of the
windings connection, therefore quite important. The zero sequence component is present in
low voltage networks, and a delta connected transformer prevents it form transferring into
higher voltage network. In this case, the unbalance factor is expressed by phase voltages
2
2 2100%A B C
U
A B C
U a U aUK
U aU a U
+ += ⋅
+ + (6.74)
where applying
92
2 1 0a a+ + = (6.75)
we obtain
2 2100%AB BC
U
AB BC
U aUK
U a U
−= ⋅
− (6.76)
and a similar expression for currents
2 2100%AB BC
I
AB BC
I aIK
I a I
−= ⋅
− (6.77)
6.5.2. Evaluation according to standards
Practically, for the computation of unbalance factor we need three phase voltages or two
phase to phase voltages, as in (6.70) to (6.77).
Another approach assumes, that the voltage vector of phase A is located on the real axis.
The phase voltages are given then as
( )
A A
j
B B
j
C C
U U
U U e
U U e
α
α β
−
− +
=
=
=
(6.78)
where α is an angle between AU and BU , and β is an angle between BU und CU . Both
angles can be computed by
2 2 2
arccos2
A B AB
A B
U U U
U Uα
+ −=
(6.79)
2 2 2
arccos2
A B AB
A B
U U U
U Uα
+ −=
(6.80)
Standards propose different formulas and levels of acceptance. In IEC 61000-2-12 [5] the
formula is given by
( )2 2 2
2
62 100%
AB BC CA
U
AB BC CA
U U UK
U U U
+ += − ⋅
+ + (6.81)
allowed levels are 2% and under special conditions 3%.
The commonly referred standard EN 50160 [66] allows the relative negative sequence value
as in (6.70) by maximally 2% of 95% of the time.
93
6.6. Literature
[1] Charles L. Fortescue, "Method of Symmetrical Co-Ordinates Applied to the Solution of
Polyphase Networks". Presented at the 34th annual convention of the AIEE
(American Institute of Electrical Engineers) in Atlantic City, N.J. on 28 July 1918.
Published in: AIEE Transactions, vol. 37, part II, pages 1027-1140 (1918)
[2] John Horak “A Derivation of Symmetrical Component Theory and Symmetrical
Component Networks”, Basler Electric Company, Presented before the59th
AnnualGeorgia Tech Protective Relaying Conference, Atlanta, Georgia, April 27-29,
2005
[3] Angelo Baggini “Handbook of Power Quality”, John Wiley &sons Ltd, The Atrium,
Southern Gate, Chichester, West Sussex PO19 8SQ. England, 2008, pp. 164-185
[4] J. Lewis Blackburn, “Symmetrical components for power systems engineering”,
Marcel Dekker Inc., 270 Madison Avenue, New York, 1993, pp. 39-55
[5] IEC 61000-2-12, Electromagnetic Compability – Part 2: Enviroment – Section 12:
Compatibility Levels for low-frequency conducted disturbances and signaling in public
medium-voltage power systems
[6] EN 50160 Voltage Characteristic of Public Distribution Systems
[7] Tadeusz Cholewicki, “Elektrotechnika Teoretyczna”, WNT, Warszawa 1973 pp. 391-
422
7. Advanced topic. Determination of transient parameters in wind
energy conversion system
The aim of the chapter is a presentation of advanced approach to the estimation of
disturbances, also assessment of power quality. A wind energy system has been chosen as
an interesting and modern network component, unfortunately producing not only green
energy but also disturbances. In this particular case, advanced signal processing methods
have been successfully applied for signal characterization and will be briefly introduced.
7.1. Motivation and problem formulation
The widespread implementation of wind energy conversion systems is a reality. Wind is seen
as a clean and renewable energy source, so the development of wind generation
94
technologies is welcomed and supported by ecologists and governments. In the next years
we will have even more generator units connected to the grid [1].
Besides distinctive merits, the connection of wind generators leads to many disturbances,
such as: voltage fluctuation, flickers, harmonics, instability, blind power regulation problems,
and transients [2]. That disturbance affects power quality. Power quality issues connected
with wind generation are not only important because of technical aspects, they are also
crucial on the free energy market.
Many of the wind energy converters installed today have a squirrel-cage induction machine
connected directly to the grid [4]. This type of the generator cannot perform voltage control
and it absorbs reactive power from the grid. Compensating capacitors are often directly
connected. That type of installation is cost-saving, and therefore still in use, but from the
system analysis point of view it can be considered as the worst case [3].
When capacitors are switched switching transients occur, whose are devastating for
sensitive equipment, protection relays and insulation. The impact of transients on power
quality indices cannot be neglected [2]. Transient overvoltages can theoretically reach peak
values up to 2.0 pu. High current transients can reach values up to ten times of the nominal
capacitor current with duration of several milliseconds [6].
The aim is the assessment of transients in an electrical system with an asynchronous
generator. The analysis is carried out for real measured signals and for different operation
conditions of the wind converter simulated in Matlab in SimPowerSystem Toolbox [5].
The Prony method was considered as an appropriate tool for parameters estimation of
transients. It shows some merits in comparison to widely applied FFT. A nonlinear regression
method was also applied and compared with Prony method. Precise computation of the
starting point of a disturbance was done using wavelets.
7.2. Prony metod
The Prony method is a technique for modelling sampled data as a linear combination of
exponential functions [6]. Although it is not a spectral estimation technique, the Prony
method has a close relationship to the least squares linear prediction algorithms used for AR
and ARMA parameter estimation. Prony method seeks to fit a deterministic exponential
model to the data in contrast to AR and ARMA methods that seek to fit a random model to
the second-order data statistics.
95
Assuming N complex data samples, the investigated function can be approximated by p
exponential functions:
( )( )1
1
[ ]+ − +
=
=∑ k k p k
pj n T j
k
k
y n A eα ω ψ
(7.1)
where
1, 2,...,=n ( , pT - sampling period, kA - amplitude, kα - damping factor, kω - angular
velocity, kψ - initial phase.
The discrete-time function may be concisely expressed in the form
1
1
[ ] −
=
=∑p
nk k
k
y n h z (7.2)
where
= kjk kh A e
ψ,
( )+= k k pj T
kz eα ω
The estimation problem is based on the minimization of the squared error over the N data
values
[ ] 21=
=∑(
n
nδ ε (7.3)
where
[ ] [ ] [ ] [ ] 1
1
−
=
= − = −∑p
nk k
k
n x n y n x n h zε (7.4)
This turns out to be a difficult nonlinear problem. It can be solved using the Prony method
that utilizes linear equation solutions.
If as many data samples are used as there are exponential parameters, then an exact
exponential fit on to the data can be made.
Consider the p-exponent discrete-time function:
1
1
[ ] −
=
=∑p
nk k
k
x n h z (7.5)
The p equations of (5) may be expressed in matrix from as:
[ ][ ]
[ ]
0 0 01 2 1
1 1 121 2
1 1 11 2
1
2
− − −
⋅ =
K
K
M MM M M
K
p
p
p p p pp
z z z h x
h xz z z
h x pz z z
(7.6)
96
The matrix equation represents a set of linear equations that can be solved for the
unknown vector of amplitudes.
Prony proposed to define the polynomial that has the exponents as its roots:
( ) ( ) ( )( ) ( )1 2
1=
= − = − − −∏ K
p
k p
k
F z z z z z z z z z (7.7)
The polynomial may be represented as the sum:
( ) [ ]
[ ] [ ] [ ] [ ]0
10 1 1
−
=
−
= =
= + + + − +
∑
K
pp m
m
p p
F z a m z
a z a z a p z a p
(7.8)
Shifting the index on (5) from n to n-m and multiplying by the parameter a[m] yield:
[ ] [ ] 1
1
[ ] − −
=
− = ∑p
n mk k
k
a m x n m a m h z (7.9)
Equation (9) can be modified into:
[ ]0
[ ] 0=
− =∑p
m
a m x n m (7.10)
The right-hand summation in (10) may be recognized as a polynomial defined by (8),
evaluated at each of its roots yielding the zero result:
[ ]
[ ]
0
1
1 0
[ ]=
− − −
= =
− =
=
∑
∑ ∑
p
m
p pn p p m
k k kk m
a m x n m
h z a m z
(7.11)
The equation can be solved for the polynomial coefficients. In the second step the roots of
the polynomial defined by (8) can be calculated. The damping factors and sinusoidal
frequencies may be determined from the roots kz .
For practical situations, the number of data points N usually exceeds the minimum number
needed to fit a model of exponentials, i.e. 2>( p . In the overdetermined data case, the
linear equation (11) should be modified to:
[ ] [ ]0
[ ]=
− =∑p
m
a m x n m e n (7.12)
The estimation problem is based on the minimization of the total squared error:
97
[ ] 21= +
= ∑r
n p
E e n (7.13)
7.3. Nonlinear regression method
Nonlinear regression is a general technique to fit a curve through data. It fits data to any
equation that defines Y as a function of X and one or more parameters. It finds the values of
those parameters that generate the curve that comes closest to the data (minimizes the sum
of the squares of the vertical distances between data points and curve). That technique
requires a model of the analyzed signal. In the case of capacitor bank switching, the signal
model of transient component can be defined as
( )1 1 1 2 2 2( ) sin sin( )−= + + +tx t A t A e tαω ϕ ω ϕ (7.14)
where A - amplitudes, α - damping factor and ω - angular velocities are unknown and
should be estimated. Practically, the signal is observed (or measured) during a finite duration
of time and ( samples of this signal are available. The measured discrete time signal
[ ]py nT can be specified as
1 1 1
( )
2 2 2
( ) sin( )
sin( ) ( )
( ) ( )
−
= + +
+ + + =
+
p
p p
nT
p p
p p
y nT A nT
A e nT e nT
x nT e nT
α
ω ϕ
ω ϕ (7.15)
for 1,2,3...=n ( .
As previously pT - is the sampling period, ( - the number of samples and e the
estimation error correlated with each sample, which includes random noise and other
distortions.
The problem of nonlinear regression can be formulated as an optimization problem, where
the goal is to minimize the difference between the physical observation and the prediction
from the mathematical model. More precisely, the goal is to determine the best values of
the unknown parameters A , α , ω , ϕ in order to minimize the squared errors between the
measured values of the signal and the computed ones. Thus, the optimization problem can
be formulated as follows:
Find a vector 1 2 1 2 1 2[ , , , , , , ] =w TA A α ω ω ϕ ϕ which minimizes the objective function
98
2 2
1 1
( ) ( ) [ ( ) ( )]= =
= = −∑ ∑w( (
p p p
n n
E e nT y nT x nT (7.16)
That is a well known standard of the least squares problem. To solve that problem, the
Quasi-Newton method was applied [7]. At each iteration, the problem is to find a new
iterate 1+kw of the form:
1+ = +w w dk k τ (7.17)
where τ is a scalar step length parameter and d is the search direction. Using the quasi-
Newton method, a line search is performed in the direction
1 ( )−= − ⋅∇d H wk kE (7.18)
where 1−H is an approximation of the inverse Hessian matrix.
The DFP formula was used for approximation of the inverse of Hessian matrix
( )
( )( )( )( )
1 1
1 11
1
− −− −+
−= + −
s s H q H qH H
q s q H q
T Tk k k k
k k
k k T Tk k k k
k
(7.19)
where ( ) ( )1+= ∇ −∇q w wk k kE E is a gradient increment and 1+= −s w wk k k is a variable
increment.
7.4. Wavelet Transform
The wavelet transform is a useful tool in signal analysis. Wavelets can be applied for the
precise computation of the beginning of a disturbing event, as it is in this paper.
The continuous Wavelet Transform (WT) of a signal ( )x t is defined as
,
1( ) ( )
+∞
−∞
−= ∫a b
t bX x t dt
aaψ (7.20)
where ( )tψ is the mother wavelet, and other wavelets
,
1( )
− =
a b
t bt dt
aaψ ψ (7.21)
are its dilated and translated versions, where a and b are the dilation parameter and
translation parameter respectively, 0, +∈ − ∈a R b R [8].
The discrete WT (DWT), instead of CWT, is used in practice [8]. Calculations are made for a
chosen subset of scales and positions. This scheme is conducted by using filters and
computing so called approximations and details. The approximations (A) are the high-scale,
99
low frequency components of the signal. The details (D) are the low-scale, high-frequency
components. The DWT coefficients are computed using the equation
, , ,[ ] [ ]∈
= =∑a b j k j k
n Z
X X x n g n (7.22)
where 2= ja , 2= jb k , ∈j ( , ∈k Z .
The wavelet filter g plays the role of ψ [8].
The decomposition (filtering) process can be iterated, so that one signal is broken down
into many lower resolution components. This is called the wavelet decomposition tree [8].
For detection of transients a multi-resolution analysis tree (Figure 7.1) based on wavelets
has been applied [9]. Each one of wavelet transform subbands is reconstructed separately
from each other, so as to get k+1 separated components of a signal x[n]. The MATLAB
multires function [10] calculates the approximation to the 2k scale and the detail signals
from the 21 to the 2k scale for a given input signal. It uses the analysis filters H (lowpass) and
G (highpass) and the synthesis filters RH and RG (lowpass and highpass respectively) (Figure
7.2).
2
2
22
2
2
2
2x[n]
2
Dx1
Dx2
Ax2
dx1
dx2
ax2
G(z)
H(z) G(z)
H(z) RH(z)
GH(z) RH(z)
RH(z)
GH(z)
Figure 7.1 . Analysis-Synthesis tree for the MATLAB multures function
As a wavelet the symlets function was used. The symlets are nearly symmetrical wavelets
proposed by Daubechies as modifications to the “db” family - orthogonal wavelets
characterized by a maximal number of vanishing moments for some given support (Figure
7.3).
100
Figure 7.2 Filters coefficients for symlets wavelet
20 30 40 50 60 70
0
0.4
0.8
20 30 40 50 60 70-1
0
1
scale
wavelet
symlet(32)
Figure 7.3 Scale and wavelet symlets function for 32 coefficients
7.5. Simulation of induction generator with capacitors
The wind generator with compensating capacitors is shown in Figure 7.4. The simulation
was done in Matlab using the SimPowerSystem Toolbox [5].
IM
TrafoGenerator
Bus Sys Bus 1 Bus 2 Bus 3
Line
Figure 7.4 Induction generator with compensating capacitors
101
A wind turbine generates power corresponding to mechanical torque on the rotating shaft of
the turbine. Produced power depends on a rotor speed and a pitch angle and is often given
in a table form [10]. The turbine characteristic used in this simulation is shown in Figure 7.5.
0 500 1000 1500 2000 2500 3000
Power [pu]
Rotor speed [rpm]
12m/s
11m/s
10m/s
9m/s
8m/s
7m/s
6m/s
0
0.2
0.4
0.6
0.8
1
1.2
1.4
Figure 7.5 Induction generator output power vs. angular velocity
The pitch control dynamic can be neglected in a power system transient analysis [4]. The
simulated generator is a 150 kW, 400 V, 1487 rpm, induction machine. It is connected to the
grid through a Dyg 25/0.4 kV distribution transformer whose nominal power was varied
between 0.5 and 2 MW during the research process and other parameters were set with
accordance to [11]. A typical 5 km overhead line [11] connected the generator to a system.
The system was represented by an equivalent source with short circuit capacity of 100 MVA
and X/R ratio of 7. The induction generator reactive power demand varies with the produced
real power [2]. During the research, different compensation levels were simulated.
Simulation results correlate with measured values.
7.6. Parameters of transients computed using Prony algorithm and
Nonlinear Regression
Firstly, to evaluate the performance of the proposed procedure simulated signals has been
used.
102
For variable wind speeds, the exact compensation of reactive power using a capacitor bank
is difficult [2]. The reactive power variation should be taken into account even if additional
capacitors are switched on and off during the operation.
Assumption of full reactive power compensation at freely chosen wind speed implies
capacitance adjustment of the compensating capacitor for every particular case. Table 7-1
contains results of the signal parameters estimation using Prony model (PRO) and nonlinear
regression (NLR). The sampling frequency was 5 kHz. Two components in the signal were
assumed and estimated: the main 50 Hz frequency [No.2] and the additional capacitor bank
switching transient [No.1]. Three wind speeds were chosen – 8 m/s, 10 m/s, 12 m/s. For
these wind speeds, the capacitor was adjusted to fully compensate reactive power. The
influence of a capacitor change on transient parameters is clearly visible (Table 7-1). For
higher wind speeds and therefore greater capacitance value, the amplitudes of the
transients were higher and the transients’ frequency lower. Time constant did not change
significantly.
Table 7-1 TRANSIENTS PARAMETERS FOR QC=80.4 KVAR AND WIND SPEEDS OF 12 M/S, 10 M/S AND 8 M/S
SignalCom./
Method
[No. –Mth.]
I
[A]
ττττ [s]
f
[Hz]
ψψψψ
[rd]
wind speed 8 m/s
1.-PRO 579.4 0.0095 526.6 -0.15
1.-NLR 579.7 0.0095 526.5 -0.14
2.-PRO 105.7 49.66 -2.37
2.-NLR 107.2 49.64 -2.36
wind speed 10 m/s
1.-PRO 624.8 0.0095 501.6 -0.15
1.-NLR 625.2 0.0095 501.6 -0.15
2.-PRO 231.1 49.82 -2.47
2.-NLR 233.3 49.81 -2.47
wind speed 12 m/s
1.-PRO 710.6 0.0095 456.5 -0.16
1.-NLR 711.1 0.0095 456.5 -0.16
2.-PRO 394.4 49.86 -2.49
2.-NLR 396.7 49.86 -2.48
Figure 7.6 shows one phase current after a capacitor switching at wind speed of 8 m/s. The
computed signals parameters are included in Table 7-1.
103
The parameter estimation using the Prony model or a nonlinear regression gave more
detailed information then Fourier spectrum estimation. Figure 7.7 depicts the spectrum of
the current shown in Figure 7.6. Two components in the signal are visible, but there is no
information about an amplitude, time constant or phase. However, Fourier method does not
need any assumption about the unknown signal model.
Apart from the capacitor and generator, also transformer, cable, short circuit level of the
system and other parameters influence the signal parameters.
Time [s]0 0.02 0.04 0.06 0.08 0.1
Current [A]
-800
-400
0
400
Figure 7.6 Current resulting from capacitor ban switching by wind speed of 08 m/s
0
20
40
60
80
100
200 400 600 800 1000 1200 1400
Current [A]
Frequency [Hz]
Figure 7.7 Fourier spectrum of the capacitor switching current
104
After getting encouraging results on simulated signals, real measured values has been
chosen for a further research activity.
Nonlinear regression and the Prony model were applied to estimate the parameters of the
measured signals.
The measurements were done on a system connected to a 500 kW induction generator with
two step compensation capacitors. A measuring recorder was connected to the high voltage
side of a 11kV/400V, Dyn11, transformer. The signal was sampled at 6250 Hz.
The most accurate detection of the beginning of transients has been achieved for 32
coefficients of symlets wavelet filters. The 1 scale detail signal enabled detection of
transients with accuracy 2-3 sampling intervals. The measured current waveform and the 1st
scale details of that signal are shown in Figure 7.8. The transients resulting from the both
switching operations were analyzed. The windowed current waveform showing the first
switching transient in detail is shown in Figure 7.9. Figure 7.10 depicts the Fourier spectrum
of the signal in Figure 7.9.
-500
0
500
1 scale detail
Current [A]
Time [s]
0 0.08 0.16 0.24 0.32 0.40
Figure 7.8 Recorded current waveform and 1 scale detail signal
105
0 0.02 0.04 0.06 0.08
Current [A]
0
400
800
1200
-400
-800
Time [s]
Figure 7.9 . Transient current after switching of the first compensating capacitor
Frequency [Hz]
500 10000
20
40
80
100
0
Current [A]
Figure 7.10 Spectrum of current by switching of first capacitor
As previously, computed signal parameters values are presented in a compact form in Table
II. The Fourier spectrum in Figure 7.10 indicates additional small components in the signal.
The signal model applied for nonlinear regression assumed only two components, so the rest
was considered to be noise.
Table 7-2 TRANSIENTS PARAMETERS
SignalCom./
Method
[No. –Mth.]
I
[A]
τ [s]
f
[Hz]
ψ
[rd]
first capacitor switching
1.-PRO 1127 0.0039 590.5 -1.80
106
1.-NLR 1123 0.0039 589.1 -1.78
2.-PRO 84.07 49.98 -0.81
2.-NLR 81.20 49.27 -0.69
second capacitor switching
1.-PRO 406 0.0032 512.2 1.57
1.-NLR 442 0.0027 512.4 1.57
2.-PRO 27.32 51.01 1.39
2.-NLR 27.15 50.83 1.44
7.7. Concluding remarks
The research results show that the Prony model and nonlinear regression are useful for
transient estimation in systems with wind generators and compensating capacitors. Those
methods enabled an accurate estimation of the amplitude, time constant, phase and
frequency of transients’ components of simulated and measured signals. The current
waveform and its parameters depend on system elements and its operating mode. Those
dependences could be observed during simulation of various cases.
The application of both Prony Model and nonlinear regression required a signal model.
Fourier transform, as a nonparametric method, did not require a signal model or even the
number of components, but could not compute signal parameters except for frequency.
Using nonlinear regression, the problem of local minima was observed. The unconsidered
setting of the initial parameters can lead to a local minimum and inadequate results.
Additionally, two signal components were predefined, so the rest was considered noise.
Application of the Prony method did not show the problem of local minima. The order of the
signal model could be easily extended to detect additional components. Both methods
delivered similar results, however, application of the Prony model seems more suitable for
an estimation of signal parameters.
7.8. Literature
[1] T. Hammons, "Status of Integrating renewable electricity production in Europe into the
grids", Proceedings Universities Power Engineering Conference, Cork, Ireland, ,p.73.
[2] Z. Lubosny, Wind Turbine Operation in Electric Power Systems, Springer Verlag, Berlin
Heidelberg (Germany), 2003, ch. 3 and 4
[3] Ch. Chompoo-inwai, L. Wei-Jen, P. Fuangfoo, M. Williams, J. R. Liao, "System Impact
107
Study for the Interconection of Wind Generation and Utility System", IEEE Trnas. on
Industry Applications, vol. 41, 2005, pp. 163-168
[4] A. Tabesh, R. Iravani, 2005, "Transient Behavior of Fixed-Speed Grid-Connected Wind
Farm", Transactions International Conference on Power Systems Transients, Paper No.
IPST05-068
[5] The Mathworks, SimPowerSystems User’s Guide, The Math Works Inc, 2000
[6] T. Lobos, J. Rezmer, H-J. Koglin, "Analysis of Power System Transients using Wavelets and
Prony Method". Proceedings IEEE Power Tech Conference, Porto, Portugal, September
2001, paper ID EMT-103
[7] A. M. Bhatti, Practical Optimization Methods. Springer Verlag, New York , USA, 2000, pp.
288-302
[8] M. Misiti, Y. Misity, G. Oppenheim, J-M. Poggi, Wavelet Toolbox User’s Guide,
MathWorks, 1996
[9] S. G. Mallat, "A theory of multiresolution signal decomposition: the wavelet
representation", IEEE Trans. Pattern Analysis and Machine Intelligence, vol. PAMI 11, July
1998, pp. 647-693
[10] S.G. Sanchez, N.G. Prelcic, S.J.G. Galan, (1996, Apr.) Uvi Wave-Wavelet Toolbox for
Matlab (ver 3.0), University of Vigo, [Online].Available:
http://www.tsc.uvigo.es/~wavelets/uvi_wave.html
[11] SIEMENS: Electrical Engineering Handbook, 1987
108
8. Computational application of power quality assessment: dips
A power quality issue includes different phenomena usually associated with deviation of the
voltage as well as current from its ideal waveform. The main aim of the activities dedicated
to power quality assessment is concentrated on a digital signal processing method
associated with general classification of power quality phenomena. Proposed in the
literature classification [1],[2],[3],[4] divides power quality in two main groups: variations
and events. The extended expression of mentioned group depicts Fig. 8.1.
Power Quality Phenomena
Variations Events
Voltage
- magnitude variation- frequency variation- unbalance- fluctuation- harmonic distortion- interharmonic component- periodic nothing- mains signaling- high frequency noise
Current
- magnitude variation- phase variation- unbalance- harmonic distortion- interharmonic component
* - note that in power quality field new types of variations and events appear regularly due to enormous range in end-user equipment, many withspecial requirements and special problems.
** - note that usually voltage events are only taken into consideration, as these can be of concern to end-user equipment. However, similarly a listof current events can be prepared.
*
Voltage
- interruption- undervoltages- magnitude steps- overvoltages- fast events- phase-angle jump- three-phase unbalance
**characteristic: small deviation around nominal or ideal value characteristic: happens every once in a while, can lead to tripping of
the equipment, to interruption of the production or of plant operation,or endanger power system reliability
Fig. 8.1 General classification of power quality phenomena
This chapter is dedicated to undervoltage event member, called voltage dips (or sags). In
standard nomenclatures dips term is associated with IEC documents and sags can be found
in IEEE Standards. Generally, a voltage dip is a reduction in the supply voltage magnitude
followed by the voltage recovery after a short period of time. Dips manifests itself in a short
duration reduction in rms voltage. The origin of the dips are usually joined with short-circuit
faults (transmission system faults, remote distribution systems faults, local distribution
system faults), starting of large motors, a short interruption and fuses. In order to localize
this event in wider group of undervoltages, some parameters should be specified: level of
the rms voltage magnitude reduction and duration time. Referring to IEC Standards, dips are
109
considered as a reduction in rms voltage to value between 90% and 1% of the declared
value, followed by a recovery between 10ms (0.5 cycles of fundamental component 50Hz)
and 1 minute later. For the IEEE Standards a voltage drop is only a sags if during a drop the
voltage is between 90% and 10%. One word of a comment requires also a phrase “declared
value”, considered as reference value for a threshold level. There are arguments in favour of
using the pre-fault voltage and there are arguments in favour of using nominal voltage. From
practical point of view, usually the base for calculation level of drop the nominal voltage is
used i.e. 0.4kV for low voltage, 10kV, 20kV for medium voltage. Classification of all voltage
magnitude events including dips definition is expressed in Fig. 8.2.
Event voltage magnitude
110%
90%
IEEE10%IEC 1%
0.5cycle10ms
Noth/tran
sient
Trans
ient
1 minute
Normal operating voltage
Voltage dip
Swell Overvoltage
Undervoltage
Event duration
Momentaryinterruption
Temporaryinterruption
3s
Fig. 8.2. Classification of all voltage magnitude events including dips definition
Taking into a count the quoted definition of voltage dips, proper classification of this event
requires an estimation of voltage magnitude as well as time duration during the dip. Thus, it
is desirable to know some advantages and disadvantages of the algorithms which allow to
detect voltage dips on the basis of recorded voltage waveform. Three approaches can be
distinguished: classical rms voltage method and peak voltage method associated with
direct time analysis and fundamental voltage component method based on frequency
analysis.
This chapter introduces mathematical backgrounds as well as a practical application of
mentioned algorithms in Matlab environment with comparison and examples.
110
8.1. Methods of voltage magnitude estimation
As it was mentioned in the introduction,there is a crucial need for permanent tracking of rms
voltage. A detection of its drop below 90% of a reference value can be treated as a threshold
for possible dip event. Among the algorithms dedicated to this task we can distinguish
methods utilizing direct sampled value (time analysis) and those required spectrum
estimation (frequency analysis). A time analysis represent such algorithm as “rms voltage
method” or “peak voltage method”. Frequency approach requires “fundamental voltage
component method”.
No mater what kind of method we select to apply tracking of the rms voltage along the
signal, the mechanism is the same – shifting executive window which leads to time trend of
a magnitude variation. It means that the main algorithm responsible for a magnitude
estimation is applied only for a particular portion of samples, designated by the window,
many times for its particular location. Finally, we achieve many local values of rms voltage
associated with a position of the window along the time axis. The representation of these
local rms values along the time axis gives desirable trend of voltage variation during the
event. Standard window width is half cycle of fundamental component for EN 50160 and
one cycle of fundamental component for IEC 61000. The trend of voltage variation can be
achieved in a high-quality mode if window is shifted sample by sample as well as in
monitoring mode when window shifts window by window with additional options of
averaging time like 200ms or 10 minutes. The described mechanism for high-quality mode is
shown in Fig. 8.3.
111
0 0.05 0.1 0.15 0.2 0.25 0.3
-1
-0.5
0
0.5
1
time[s]
volta
ge [p.u.]
k=1
v(1)v(2)...
V(N)
Vrms(1)
k=1
v(1)v(2)...
V(N)
Vrms(M-N)
In high-quality mode the window is shifted sample
by sample
M - total number of samples
N - number of samples take into calculation of local rms value/number of samples of the window
M-N - number of samples of voltage magnitude characteristic inhigh-quality mode
0.05 0.1 0.15 0.2 0.25 0.3
0.7
0.75
0.8
0.85
0.9
0.95
1Example of rms voltage magnitude during sags for one-cycle (20ms) window
volta
ge m
agn
itude
[p.u.]
Vrms(1)
0.02
Vrms(M-N)
0.3398
0.3398time[s]
local application of
the algorithm
Fig. 8.3. Mechanism of rms voltage trend during the event: high-quality mode, one-cycle algorithm
“rms voltage” method
This method is based on the classical definition of rms value of periodical signals
[1],[3],[4],[5]:
( ) ( )1
211
i k (
i k
Vrms k v i(
= + −
=
= ∑ (8.1)
where: k =(1,2…M-N), M – the total length of an investigated voltage signal v in samples,
i=(k,k+1…k+N-1) – indicates a set of samples v(i) taken into calculation of local Vrms1(k),
always on the basis of N samples, N – a number of samples taken for calculation – usually
equals an integer multiplication of a number of samples per one cycle of fundamental
component.
Note, we obtain that first value of estimated rms voltage Vrms1(1) on the basis of first N
samples. Thus, following by the time axis the first value of estimated rms voltage is localized
in time moment equals N*dt where dt is a sampling step. The next points of desirable
112
Vrms1(k) are calculated using shifting window. For a high quality calculation the shift is fine,
that means sample by sample. In monitoring quality mode the rms value is calculated once
per cycle or half-cycle. Additionally, the averaging time is introduced concerning one value
for selected numbers of cycles (half-cycles) i.e. 200ms averaging time means that we achieve
one value from 10 one-cycle values or 20 half-cycles values.
“peak voltage” method
Similarly to the previous approach, we expect to achieve a function of time that corresponds
to rms voltage changes during the events. Here for each samples over the proceeding
window of the voltage, the maximum of absolute value is calculated. Again, the length of
executive window should contain integer multiplication of a number of samples per one
cycle of fundamental component. Finally, we can estimate rms voltage for particular
localisation of the window by [1],[3],[4],[5]:
( )( ) max
2 , ( , 1,..., 1)2
v iVrms k i k k k (= = + + − (8.2)
where: k =(1,2…M-N), M – the total length of investigated voltage signal v in samples,
i=(k,k+1…k+N-1) – indicates a set of samples v(i) taken into calculation of local Vrms2(k),
always on the basis of N samples, N – a number of samples taken for calculation – usually
equals an integer multiplication of a number of samples per one cycle of fundamental
component.
Similarly, followed by the time axis first value of estimated rms voltage is localized in time
moment equals N*dt where dt is the sampling step. The next points of desirable Vrms2(k)
are calculated using the shifting window till the end of recorded data. For a high quality
calculation the shift is fine, which means sample by sample shift. In a monitoring quality
mode the rms value is calculated once per cycle or per half-cycle.
“fundamental voltage component” method
This approach introduces a frequency domain with a classical method for spectrum
estimation method that Fourier analysis is. In a continuous-time domain we can calculate
Fourier transform and inverse Fourier transform in order to obtain a desirable fundamental
component f0:
113
( ) ( ) ( ) ( )0
0
1,
2
j t j tv t V j e d V j v t e dt
ω ω
ω
ω ω ωπ
+∞ +∞−
− −∞
= ⋅ ⇐ = ⋅∫ ∫ (8.3)
In a discrete time-domain, we reintroduce this approach applying Fast Fourier Transform
(FFT). The idea of tracking rms voltage in time is equally preserved compared to previous
method i.e. a local value of rms voltage are associated with a particular position of executive
window along time axis. However, the algorithm of local rms voltage utilizes frequency
approach this time [1],[3],[4],[5]:
( ) ( )( )50
3( ) ( ) , , 1,.., 1
50 50 / 50 / /
m mVrms k V m V m FFT v i i k k k (
m df fs (
== ⇐ = = + + −
= =
(8.4)
where: k =(1,2…M-N), M – the total length of investigated voltage signal v in samples,
i=(k,k+1…k+N-1) – indicates a set of samples v(i) taken into calculation of local Vrms3(k),
always on the basis of N samples, N – a number of samples taken for calculation – usually
equals an integer multiplication of number of samples per one cycle of fundamental
component.
On the basis of local N samples of a windowed signal v(i) the FFT is performed. It transfers
the set of local v(i) data into a frequency domain represented now by a local vector V(m).
Note that V(m) contains a complex value. The next step is to select from this vector an
element associated with fundamental 50Hz component. In order to do it, we have to
recalculate the index m which represents 50Hz on the basis of sampling frequency (fs) and a
number of frequency bins, here equals N. Defining frequency resolution as df=fs/N, we can
uncover an index associated with fundamental component m50=50/df. Finally, we identify
complex vector describing magnitude as well as phase of fundamental component as
V(m50).
Even though we use a frequency approach, we still intend to track the rms voltage in time.
Thus, followed by the time axis, the first value of estimated rms voltage is localized in time
moment equals N*dt where dt is the sampling step. The next points of desirable Vrms3(k)
are calculated using a shifting window till the end of recorded data. For a high quality
calculation the shift is fine, which means sample by sample shift. In a monitoring quality
mode the rms value is calculated once per cycle.
Warning: Applying FFT algorithm, we have to be conscious of some limitations. This
algorithm gives desirable effect if the number of samples taken for FFT calculation equals
integer value of samples per one cycle of fundamental component. If not, the leakage effect
114
must be considered. Thus, application of this approach in case of half-cycle window is more
complicated. One of the proposed solution is to take a half-cycle window and build second
half-cycle window as:
( ) ( ) ( ) ( ) ( ) ( ) ( ) , 1 ,..., 1 , , 1 ,..., 1 ,v i v k v k v k ( v k v k v k ( only for ( half cycle window= + + − − − + − + − = −
(8.5)
8.2. Application of rms trend calculation in Matlab
Reconnaissance in a signal processing area manifest itself in crucial need for some details
about the origin of the phenomena (physical sphere) and parameters of quantisation,
especially a sampling rate (digital sphere).
Before we start working with analysis of the phenomena, it is necessary to perform some
initial calculations concerning sampling rate, definition of time axis etc.
This section demonstrates investigations of real measured voltage phenomena during
decaying fault in low voltage distribution systems under a storm condition Fig. 8.4. File is
also available in Matlab format at http://eportal.eny.pwr.wroc.pl/ as sag_dg.mat. From the
digital point of view sampling rate is 10.240kHz. Vectors consist of 4096 samples. In order to
demonstrate the phenomena and to associate time axis with recording parameters some
initial calculation should be performed.
Initial calculation:
sampling rate (sampling frequency) sf 10 240kHz= .
time step (time resolution) 9.7656e-005s
1 1dt s
f 10240= = =
time axis [ ] [ ]t 0 1 4096 1 dt 0 dt 0 4 dt s= − = −: : * : : .
115
0 0.05 0.1 0.15 0.2 0.25 0.3 0.35
-300
-200
-100
0
100
200
300
Voltage sag during grid disturbances
time [s]
volta
ge [V]
ul1
uL2
uL3
Fig. 8.4. Visualisation of example of investigated dip phenomena
The aim of this chapter is to present some example algorithms in Matlab environment and
results concerning discussed methods of rms calculation as well as the influence of the
window width.
Table 8-1. Demonstration algorithm of rms trend calculation using classical “rms voltage”
method
Fig. 8.5. Investigated one-phase signal and its rms trend characteristic using half-cycle “rms
voltage” method
Table 8-2. Demonstration algorithm of rms trend calculation using “peak voltage” method
Fig. 8.6. Investigated one-phase signal and its rms trend characteristic using half-cycle “peak
voltage” method.
Table 8-3. Demonstration algorithm of rms trend calculation using “fundamental voltage
component” method
Fig. 8.7. Investigated one-phase signal and its rms trend characteristic using half-cycle
“fundamental voltage component” method
116
Table 8-1. Demonstration algorithm of rms trend calculation using classical “rms voltage” method
%---------SIGNAL------------------------------------------------------------- clear all, close all load sag_gd; %sag_gd.mat contain uL1 uL2 uL3 4096 samples %fs=10240Hz %it covers 20 cycles (period of fundamental component) %4096 total number of samples %2048 samples/ 10cycle %204.8 samples/ 1cycle, round 204 sample/cycle x=uL1; h=plot(t,x) set(h,'LineWidth',2); xlabel('time [s]') ylabel('voltage [V]') grid on axis('tight') %---------INITIAL PARAMETERS-------------------------------------------------------------- icycle=1/2; %icycle- executive window width in cycles of fundamental component fo=50; %fundamental component 50Hz To=1/fo; %one cycle of the fundamental component in [s] tN=icycle*To; %executive window width in [s] N=round(icycle*To/dt) %executive window width in [samples] %----------------------------------------------------------------------- %----------------------------------------------------------------------- %TIME ANALYSIS - Vrms1- rms voltage: 1-phase, high quality %----------------------------------------------------------------------- %Vrms1 - vector of local rms voltage %calculated on the basis of N samples of the signal %High quality mode - every next Vrms1 are obtained %by shifting the window along signal sample by sample. %Note - first value of Vrms1 can be obtained after recording of %assumed N samples of executive window (tN in seconds). %Thus we should rescale time axis in Figure representing rms voltage %starting from tN and finish with end of the signal ->t1 %For high quality mode we preserve time step in t1 as dt for k=1:length(x)-N k n=[k:k+N-1]; %identification of windowed samples xN=[x(n)]; %windowed signal samples taken for calculation of local rms Vrms1(k)=sqrt(1/N*sum(xN.^2)); end t1=t(N+1:length(x)); figure h=plot(t1,Vrms1) set(h,'LineWidth',2) xlabel('time [s]') ylabel('voltage magnitude [V]') title(['Vrms1: ',num2str(icycle),'-cycle window']) grid on axis('tight') %-----------------------------------------------------------------------
0 0.05 0.1 0.15 0.2 0.25 0.3 0.35
-300
-200
-100
0
100
200
300
time [s]
volta
ge [V]
0.05 0.1 0.15 0.2 0.25 0.3 0.35
100
120
140
160
180
200
220
240
time [s]
volta
ge m
agnitude
[V]
Vrms1: 0.5-cycle window
Fig. 8.5. Investigated one-phase signal and its rms trend characteristic using half-cycle “rms voltage” method
117
Table 8-2. Demonstration algorithm of rms trend calculation using “peak voltage” method
%---------SIGNAL------------------------------------------------------------- clear all, close all load sag_gd; %sag_gd.mat contain uL1 uL2 uL3 4096 samples %fs=10240Hz %it covers 20 cycles (period of fundamental component) %4096 total number of samples %2048 samples/ 10cycle %204.8 samples/ 1cycle, round 204 sample/cycle x=uL1; h=plot(t,x) set(h,'LineWidth',2); xlabel('time [s]') ylabel('voltage [V]') grid on axis('tight') %---------INITIAL PARAMETERS-------------------------------------------------------------- icycle=1/2; %icycle- executive window width in cycles of fundamental component fo=50; %fundamental component 50Hz To=1/fo; %one cycle of the fundamental component in [s] tN=icycle*To; %executive window width in [s] N=round(icycle*To/dt) %executive window width in [samples] %----------------------------------------------------------------------- %----------------------------------------------------------------------- %TIME ANALYSIS - Vrms2 - peak voltage: 1-phase, high quality %----------------------------------------------------------------------- %Vrms2 - vector of local rms voltage %calculated on the basis of N samples of the signal %High quality mode - every next Vrms2 are obtained %by shifting the window along signal sample by sample. %Note - first value of Vrms2 can be obtained after recording of %assumed N samples of executive window (tN in seconds). %Thus we should rescale time axis in Figure representing rms voltage %starting from tN and finish with end of the signal ->t1 %For high quality mode we preserve time step in t1 as dt for k=1:length(x)-N k n=[k:k+N-1]; %identification of windowed samples xN=[x(n)]; %windowed signal samples taken for calculation of local rms Vrms2(k)=1/sqrt(2).*(max(abs(x(n)))); end t2=t(N+1:length(x)); figure h=plot(t2,Vrms2) set(h,'LineWidth',2) xlabel('time [s]') ylabel('voltage magnitude [V]') title(['Vrms2: ',num2str(icycle),'-cycle window']) grid on axis('tight')
0 0.05 0.1 0.15 0.2 0.25 0.3 0.35
-300
-200
-100
0
100
200
300
time [s]
volta
ge [V]
0.05 0.1 0.15 0.2 0.25 0.3 0.35
100
150
200
250
time [s]
volta
ge m
agnitude
[V]
Vrms2: 0.5-cycle window
Fig. 8.6. Investigated one-phase signal and its rms trend characteristic using half-cycle “peak voltage” method.
118
Table 8-3. Demonstration algorithm of rms trend calculation using “fundamental voltage component”
method
%---------SIGNAL------------------------------------------------------------- clear all, close all load sag_gd; %sag_gd.mat contain uL1 uL2 uL3 4096 samples %fs=10240Hz %it covers 20 cycles (period of fundamental component) %4096 total number of samples %2048 samples/ 10cycle %204.8 samples/ 1cycle, round 204 sample/cycle x=uL1; h=plot(t,x) set(h,'LineWidth',2); xlabel('time [s]') ylabel('voltage [V]') grid on axis('tight') %---------INITIAL PARAMETERS-------------------------------------------------------------- icycle=1/2; %icycle- executive window width in cycles of fundamental component fo=50; %fundamental component 50Hz To=1/fo; %one cycle of the fundamental component in [s] tN=icycle*To; %executive window width in [s] N=round(icycle*To/dt) %executive window width in [samples] %----------------------------------------------------------------------- %----------------------------------------------------------------------- %FREQUENCY ANALYSIS - Vrmsfundamental voltage componnent %----------------------------------------------------------------------- switch icycle case 0.5 p=To/dt; %p-delta phase shift m=[0:N*2-1]; %frequency axis in [binNnumber] fanalysis=(fs*m)/(N*2); %frequency axis in [Hz] m50=round(50*(N*2)/fs+1); %bin of 50Hz for k=1:(length(x)-N); k n=[k:k+N-1]; %identification of windowed samples xN=[x(n),-x(n)]; %special series windowed signal samples taken for FFT localspectrumu=fft(xN,N*2); %local complex rms voltage of fundamental component fundamental50uc=localspectrumu(m50)*exp(-j*(2*pi/p)*(k-1))./(sqrt(2)); s1=2/(N*2); %scale factor for magnitude real50u(k)=real(fundamental50uc).*s1; %real part of local rms voltage of fundamental component img50u(k)=imag(fundamental50uc).*s1; %imaginary part of local rms voltage of fundamental component s2=180/pi;%scale factor for phase in angle phase50u(k)=(angle(fundamental50uc)+pi/2)*s2; %local phase of fundamental component as sin end Vrms3=sqrt((real50u).^2+(img50u).^2); t3=t(N+1:length(x)); otherwise p=To/dt; %p-delta phase shift m=[0:N-1]; %frequency axis in [bin number] fanalysis=(fs*m)/(N); %frequency axis in [Hz] m50=round(50*(N)/fs+1); %bin of 50Hz for k=1:(length(x)-N); k n=[k:k+N-1]; %identification of windowed samples xN=[x(n)]; %windowed signal samples taken for FFT localspectrumu=fft(xN,N); %local complex rms voltage of fundamental component fundamental50uc=localspectrumu(m50)*exp(-j*(2*pi/p)*(k-1))./(sqrt(2)); s1=2/(N); %scale factor for magnitude real50u(k)=real(fundamental50uc).*s1; %real part of local rms voltage of fundamental component img50u(k)=imag(fundamental50uc).*s1; %imaginary part of local rms voltage of fundamental component s2=180/pi;%scale factor for phase in angle phase50u(k)=(angle(fundamental50uc)+pi/2)*s2; %local phase of fundamental component as sin end Vrms3=sqrt((real50u).^2+(img50u).^2); t3=t(N+1:length(x)); end figure h=plot(t3,Vrms3) set(h,'LineWidth',2) xlabel('time [s]') ylabel('voltage magnitude [V]') title(['Vrms3: ',num2str(icycle),'-cycle window']) grid on axis('tight') %-----------------------------------------------------------------------
119
0 0.05 0.1 0.15 0.2 0.25 0.3 0.35
-300
-200
-100
0
100
200
300
time [s]
volta
ge [V]
0.05 0.1 0.15 0.2 0.25 0.3 0.35
100
120
140
160
180
200
220
240
time [s]
volta
ge m
agnitude
[V]
Vrms3: 0.5-cycle window
Fig. 8.7. Investigated one-phase signal and its rms trend characteristic using half-cycle “fundamental voltage
component” method
120
8.3. Comparison of the methods
The comparison of the methods used in process of the rms calculation can be performed on
the basis of dynamism of obtained characteristic and computational power and time
consumed for calculation. Fig. 8.8 indicates the “peak value” method as a rough estimation
of rms changes. This method is the fastest for calculation. The algorithm based on classical
rms calculation is comparable with “fundamental component”, however a classical operation
does not involve frequency domain calculation and thereby is more attractive in the point of
the fast practical application.
0.05 0.1 0.15 0.2 0.25 0.3 0.35
100
150
200
250
time [s]
volta
ge m
agnitude
[V]
Vrms1,Vrms2,Vrms3: 0.5-cycle window
Vrms1
Vrms2
Vrms3
Fig. 8.8. Comparison of the “rms voltage” method (Vrms1), “peak-voltage” method (Vrms2) and “fundamental
component” method (Vrms3) for given half-cycle window
The window width, usually defined as a number of cycles of fundamental component, has a
crucial meaning for dynamism of rms characteristic, no matter what kind of method is
applied. Fig. 8.9,Fig. 8.10,Fig. 8.11 depict a common effect of smearing and reduction of the
dynamism of the characteristic when windows longer than one period of fundamental
component is applied. Thus, according to the standards, a recommended base of the
algorithm is half or one-period of fundamental component.
121
0.05 0.1 0.15 0.2 0.25 0.3 0.35
100
120
140
160
180
200
220
240
time [s]
volta
ge m
agnitude
[V]
Vrms1: 0.5,1,2-cycle window
0.5-cycle
1-cycle
2-cycle
Fig. 8.9. Influence of the window width on “rms voltage” method
0.05 0.1 0.15 0.2 0.25 0.3 0.35
100
150
200
250
time [s]
volta
ge m
agnitude
[V]
Vrms2: 0.5,1,2-cycle window
0.5-cycle
1-cycle
2-cycle
Fig. 8.10. Influence of the window width on “peak voltage” method
122
0.05 0.1 0.15 0.2 0.25 0.3 0.35
100
120
140
160
180
200
220
240
time [s]
volta
ge m
agnitude
[V]
Vrms3: 0.5,1,2-cycle window
0.5-cycle
1-cycle
2-cycle
Fig. 8.11. Influence of the window width on “fundamental component” method
8.4. Dip duration
Previous section was concentrated on voltage magnitude estimation during the event. The
dip duration is strongly associated with the magnitude variation as a time interval when
voltage magnitude is below a given threshold. Referring to the Standards, the threshold
initiating dip recognition is 90% of a reference value. Having the detailed information about
the recorded phenomena, its origin and voltage condition, we can select nominal value as
the reference. However, there are the opinions that allow adapting measured voltage in
steady state, before the event, as a reference value. From practical point of view suggested
approach uses nominal voltage as reference value in order to set up the threshold.
An example of dip duration identification on the basis of one-cycle “rms voltage” algorithm
is shown in Fig. 8.12. The reference value is the nominal voltage in per unit value and the
reshold is setup on 90%.
123
0.02 0.04 0.06 0.08 0.1 0.110.2
0.3
0.4
0.5
0.6
0.7
0.8
0.9
1
time [s]
Vrms1: 1-cycle window
Threshold: 90%
0.0904s0.0267s
Sag duration=0.0904-0.0267=0.067s (3.1836 cycles)
Fig. 8.12. Voltage dip duration identification
Again it must be emphasised that we stay in the high quality mode calculation. It means that
estimated voltage magnitude is updated every sample. For a monitoring quality rms value
will be calculated once per averaging time on the basis of selected number of half or one-
cycle values.
In order to explore the dip duration identification, it is suggested to search automatically the
output vectors of rms characteristic in order to indicate values which do not fulfil a given
threshold. These values stay in reference to the time axis and indicate dip initiation and
voltage recovery. The first activity uncovers influence of the threshold level on dip duration
obtained thanks to three applied rms voltage estimation algorithm for given one-cycle
window. Second activity aims to compare obtained dip duration for one-cycle window with
given threshold 90%. Proposed approach can be treated as an academic off-line calculation
in order to uncover the idea of the dip identification using the threshold. Normally, in on-line
application, this identification is done automatically, in real time, on the basis of every new
sample which does not fulfil requirements for a given threshold. Presented in this section
attitude serves as comparison of the threshold level on detected duration time of the dip.
Table 8-4 gives a proposition of demonstration algorithm in Matlab dedicated to an
influence of a threshold on dip duration. The effects of comparison is presented in Fig. 8.13.
124
Table 8-4. Demonstration algorithm applying discussed method and comparison of threshold level on dip
duration
%---------SIGNAL------------------------------------------------------------- clear all, close all load sag_gd; %sag_gd.mat contain uL1 uL2 uL3 4096 samples %fs=10240Hz %it covers 20 cycles (period of fundamental component) %4096 total number of samples %2048 samples/ 10cycle %204.8 samples/ 1cycle, round 204 sample/cycle x=uL1; h=plot(t,x) set(h,'LineWidth',2); xlabel('time [s]') ylabel('voltage [V]') grid on axis('tight') %---------INITIAL PARAMETERS-------------------------------------------------------------- icycle=1/2; %icycle- executive window width in cycles of fundamental component fo=50; %fundamental component 50Hz To=1/fo; %one cycle of the fundamental component in [s] tN=icycle*To; %executive window width in [s] N=round(icycle*To/dt) %executive window width in [samples] %----------------------------------------------------------------------- %----------------------------------------------------------------------- %TIME ANALYSIS - Vrms1- rms voltage: 1-phase, high quality %----------------------------------------------------------------------- %Vrms1 - vector of local rms voltage %calculated on the basis of N samples of the signal %High quality mode - every next Vrms1 are obtained %by shifting the window along signal sample by sample. %Note - first value of Vrms1 can be obtained after recording of %assumed N samples of executive window (tN in seconds). %Thus we should rescale time axis in Figure representing rms voltage %starting from tN and finish with end of the signal ->t1 %For high quality mode we preserve time step in t1 as dt for k=1:length(x)-N k n=[k:k+N-1]; %identification of windowed samples xN=[x(n)]; %windowed signal samples taken for calculation of local rms Vrms1(k)=sqrt(1/N*sum(xN.^2)); end t1=t(N+1:length(x)); figure h=plot(t1,Vrms1) set(h,'LineWidth',2) xlabel('time [s]') ylabel('voltage magnitude [V]') title(['Vrms1: ',num2str(icycle),'-cycle window']) grid on axis('tight') %----------------------------------------------------------------------- %----------------------------------------------------------------------- %TIME ANALYSIS - Vrms2 - peak voltage: 1-phase, high quality %----------------------------------------------------------------------- %Vrms2 - vector of local rms voltage %calculated on the basis of N samples of the signal %High quality mode - every next Vrms2 are obtained %by shifting the window along signal sample by sample. %Note - first value of Vrms2 can be obtained after recording of %assumed N samples of executive window (tN in seconds). %Thus we should rescale time axis in Figure representing rms voltage %starting from tN and finish with end of the signal ->t1 %For high quality mode we preserve time step in t1 as dt for k=1:length(x)-N k n=[k:k+N-1]; %identification of windowed samples xN=[x(n)]; %windowed signal samples taken for calculation of local rms Vrms2(k)=1/sqrt(2).*(max(abs(x(n)))); end t2=t(N+1:length(x)); figure h=plot(t2,Vrms2) set(h,'LineWidth',2) xlabel('time [s]') ylabel('voltage magnitude [V]') title(['Vrms2: ',num2str(icycle),'-cycle window']) grid on axis('tight')
125
%----------------------------------------------------------------------- %FREQUENCY ANALYSIS - fundamental voltage componnent %----------------------------------------------------------------------- switch icycle case 0.5 p=To/dt; %p-delta phase shift m=[0:N*2-1]; %frequency axis in [bin number] fanalysis=(fs*m)/(N*2); %frequency axis in [Hz] m50=50*(N*2)/fs+1; %bin of 50Hz for k=1:(length(x)-N); k n=[k:k+N-1]; %identification of windowed samples xN=[x(n),-x(n)]; %special series windowed signal samples taken for FFT localspectrumu=fft(xN,N*2); %local complex rms voltage of fundamental component fundamental50uc=localspectrumu(m50)*exp(-j*(2*pi/p)*(k-1))./(sqrt(2)); s1=2/(N); %scale factor for magnitude real50u(k)=real(fundamental50uc).*s1; %real part of local rms voltage of fundamental component img50u(k)=imag(fundamental50uc).*s1; %imaginary part of local rms voltage of fundamental component s2=180/pi;%scale factor for phase in angle phase50u(k)=(angle(fundamental50uc)+pi/2)*s2; %local phase of fundamental component end Vrms3=sqrt((real50u).^2+(img50u).^2); t3=t(N+1:length(x)); otherwise p=To/dt; %p-delta phase shift m=[0:N-1]; %frequency axis in [bin number] fanalysis=(fs*m)/(N); %frequency axis in [Hz] m50=50*(N)/fs+1; %bin of 50Hz for k=1:(length(x)-N); k n=[k:k+N-1]; %identification of windowed samples xN=[x(n)]; %windowed signal samples taken for FFT localspectrumu=fft(xN,N); %local complex rms voltage of fundamental component fundamental50uc=localspectrumu(m50)*exp(-j*(2*pi/p)*(k-1))./(sqrt(2)); s1=2/(N); %scale factor for magnitude real50u(k)=real(fundamental50uc).*s1; %real part of local rms voltage of fundamental component img50u(k)=imag(fundamental50uc).*s1; %imaginary part of local rms voltage of fundamental component s2=180/pi;%scale factor for phase in angle phase50u(k)=(angle(fundamental50uc)+pi/2)*s2; %local phase of fundamental component end Vrms3=sqrt((real50u).^2+(img50u).^2); t3=t(N+1:length(x)); end figure h=plot(t3,Vrms3) set(h,'LineWidth',2) xlabel('time [s]') ylabel('voltage magnitude [V]') title(['Vrms3: ',num2str(icycle),'-cycle window']) grid on axis('tight') %----------------------------------------------------------------------- %----------------------------------------------------------------------- vartreshold=[0.95 0.9 0.85 0.8]; %various treshold level for given icycle %----------------------------------------------------------------------- for m=1:4 treshold=vartreshold(m) durationVrms1samples=find(Vrms1<=treshold*Vrms1(1)); beginVrms1=t1(durationVrms1samples(1)) endVrms1=t1(durationVrms1samples(length(durationVrms1samples))) durationVrms1seconds(m)=endVrms1-beginVrms1 durationVrms1cycles(m)=durationVrms1seconds(m)/To durationVrms2samples=find(Vrms2<=treshold*Vrms2(1)); beginVrms2=t2(durationVrms2samples(1)) endVrms2=t2(durationVrms2samples(length(durationVrms2samples))) durationVrms2seconds(m)=endVrms2-beginVrms2 durationVrms2cycles(m)=durationVrms2seconds(m)/To durationVrms3samples=find(Vrms3<=treshold*Vrms3(1)); beginVrms3=t3(durationVrms3samples(1)) endVrms3=t3(durationVrms3samples(length(durationVrms3samples))) durationVrms3seconds(m)=endVrms3-beginVrms3 durationVrms3cycles(m)=durationVrms3seconds(m)/To end h=plot(vartreshold,durationVrms1seconds,'--bs','MarkerEdgeColor','k','MarkerFaceColor','b','MarkerSize',8) set(h,'LineWidth',2) hold on h=plot(vartreshold,durationVrms2seconds,'--gv','MarkerEdgeColor','k','MarkerFaceColor','g','MarkerSize',8) set(h,'LineWidth',2) hold on h=plot(vartreshold,durationVrms3seconds,'--rd','MarkerEdgeColor','k','MarkerFaceColor','r','MarkerSize',8) set(h,'LineWidth',2) hold off xlabel('voltage treshold') ylabel('dip duration [s]') title(['Vrms1,Vrms2,Vrms3: ',num2str(icycle),'-cycle window']) legend('Vrms1','Vrms2','Vrms3') grid on axis('tight') %-----------------------------------------------------------------------
126
0 0.05 0.1 0.15 0.2 0.25 0.3 0.35
-300
-200
-100
0
100
200
300
time [s]
volta
ge [V]
0.8 0.85 0.9 0.95
0.17
0.175
0.18
0.185
0.19
0.195
0.2
voltage treshold
dip du
ratio
n [s]
Vrms1,Vrms2,Vrms3: 0.5-cycle window
Vrms1
Vrms2
Vrms3
Fig. 8.13. Investigated one-phase signal and comparison of threshold level on dip duration
8.5. Phase-angle jump
In power systems voltage and current is a complex quantity (phasor) which has magnitude
and phase. The description of such event as dips is not limited to magnitude variation but
includes also phase changes. It has a meaning when we work with sensitive equipment as
power converters which use phase-angle information for their firing instant. The phase-angel
jump manifests itself as shift in zero-crossing of the instantaneous voltage. In order to define
a phase-angle jump, we need to track a difference between a phase-angle before the event
(steady state) and a phase-angle during the event. The instantaneous value of the voltage
can be calculated using voltage zero-crossing method (time analysis) or from the phase of
fundamental component (frequency analysis).
As it was mentioned in the previous section, the voltage “fundamental component” method
introduced frequency analysis using FFT application. In a frequency domain, particular
components are the complex value represented by magnitude and phase.
In a digital application, we have to consider one additional issue. In a continuous time-
domain the instantaneous voltage waveform of fundamental component is considered as a
rotating phasor. The rotation is described by rotation operator and consists two
components: time operator 0exp( j t )ω− and phase operator exp( j )ψ− :
( ) ( ) ( ) 0 0
0 0
j( t ) j t j
maxv t V sin t V t V e V e e
ω ψ ω ψω ψ += ⋅ + ⇒ = ⋅ = ⋅ ⋅ (8.6)
In the point of phase tracking the most important is the phase operator. However, in the
digital application using the shifting window, we introduce a rotation operator as complex
127
mechanism. Thus, in order to identify desirable phase operator we have to “move back” the
phasor by correction 0
exp( j t )ω− . In a digital notice, it means exp(-j*(2*pi/p)*(k-1) where
p=T0/dt, dt – sampling rate, T0 – period of fundamental component.
Table 8-5 contains a demonstration algorithm in Matlab realising a phase-angle jump
assessment on the basis of fundamental phasor changes. The transition of phase-angle
changes is presented in Fig. 8.14. Additionally, the influence of window width on discussed
characteristic is shown in Fig. 8.15.
128
Table 8-5. Demonstration algorithm applying fundamental component method for phase-angle jump
calculation
%---------SIGNAL------------------------------------------------------------- clear all, close all load sag_gd; %sag_gd.mat contain uL1 uL2 uL3 4096 samples %fs=10240Hz %it covers 20 cycles (period of fundamental component) %4096 total number of samples %2048 samples/ 10cycle %204.8 samples/ 1cycle, round 204 sample/cycle x=uL1; h=plot(t,x) set(h,'LineWidth',2); xlabel('time [s]') ylabel('voltage [V]') grid on axis('tight') %---------INITIAL PARAMETERS-------------------------------------------------------------- icycle=1/2; %icycle- executive window width in cycles of fundamental component fo=50; %fundamental component 50Hz To=1/fo; %one cycle of the fundamental component in [s] tN=icycle*To; %executive window width in [s] N=round(icycle*To/dt) %executive window width in [samples] %----------------------------------------------------------------------- %----------------------------------------------------------------------- %FREQUENCY ANALYSIS - Vrmsfundamental voltage componnent %----------------------------------------------------------------------- switch icycle case 0.5 p=To/dt; %p-delta phase shift m=[0:N*2-1]; %frequency axis in [bin number] fanalysis=(fs*m)/(N*2); %frequency axis in [Hz] m50=round(50*(N*2)/fs+1); %bin of 50Hz for k=1:(length(x)-N); k n=[k:k+N-1]; %identification of windowed samples xN=[x(n),-x(n)]; %special series windowed signal samples taken for FFT localspectrumu=fft(xN,N*2); %local complex rms voltage of fundamental component s1=2/(N*2); %scale factor for magnitude s2=180/pi;%scale factor for phase in angle fundamental50uc=localspectrumu(m50)*exp(-j*(2*pi/p)*(k-1))./(sqrt(2)).*s1; real50u(k)=real(fundamental50uc); %real part of local rms voltage of fundamental component img50u(k)=imag(fundamental50uc); %imaginary part of local rms voltage of fundamental component s2=180/pi;%scale factor for phase in angle phase50u(k)=(angle(fundamental50uc)+pi/2)*s2; %local phase of fundamental component as sin end Vrms3=sqrt((real50u).^2+(img50u).^2); t3=t(N+1:length(x)); otherwise p=To/dt; %p-delta phase shift m=[0:N-1]; %frequency axis in [bin number] fanalysis=(fs*m)/(N); %frequency axis in [Hz] m50=round(50*(N)/fs+1); %bin of 50Hz for k=1:(length(x)-N); k n=[k:k+N-1]; %identification of windowed samples xN=[x(n)]; %windowed signal samples taken for FFT localspectrumu=fft(xN,N); %local complex rms voltage of fundamental component s1=2/(N); %scale factor for magnitude s2=180/pi;%scale factor for phase in angle fundamental50uc=localspectrumu(m50)*exp(-j*(2*pi/p)*(k-1))./(sqrt(2)).*s1; real50u(k)=real(fundamental50uc); %real part of local rms voltage of fundamental component img50u(k)=imag(fundamental50uc); %imaginary part of local rms voltage of fundamental component phase50u(k)=(angle(fundamental50uc)+pi/2)*s2; %local phase of fundamental component as sin end Vrms3=sqrt((real50u).^2+(img50u).^2); t3=t(N+1:length(x)); end %----------------------------------------------------------------------- figure h=plot(t3,Vrms3) set(h,'LineWidth',2) xlabel('time [s]') ylabel('voltage magnitude [V]') title(['Vrms3: ',num2str(icycle),'-cycle window']) grid on axis('tight') figure h=plot(t3,phase50u) set(h,'LineWidth',2) xlabel('time [s]') ylabel('phase [deg]') title(['phase: ',num2str(icycle),'-cycle window']) grid on axis('tight') phase50uanglejump=phase50u-phase50u(1); figure h=plot(t3,phase50uanglejump) set(h,'LineWidth',2) xlabel('time [s]') ylabel('phase-angle jump [deg]') title(['phase-angle jump: ',num2str(icycle),'-cycle window']) grid on axis('tight')
129
0 0.05 0.1 0.15 0.2 0.25 0.3 0.35
-300
-200
-100
0
100
200
300
time [s]
volta
ge [V]
0.05 0.1 0.15 0.2 0.25 0.3 0.35
-30
-20
-10
0
10
20
30
time [s]
phas
e-an
gle jump [deg
]
phase-angle jump: 0.5-cycle window
Fig. 8.14. Investigated one-phase signal and its phase-angle jump
0.05 0.1 0.15 0.2 0.25 0.3 0.35
-30
-20
-10
0
10
20
30
time [s]
phas
e an
gle jump [deg
]
Vrms3: 0.5,1,2-cycle window
0.5-cycle
1-cycle
2-cycle
Fig. 8.15. Influence of the window width on on transition of phase-angel jump characteristic
8.6. Monitoring mode in voltage dips assessment
The previous sections described three algorithms dedicated to voltage magnitude
deterioration: rms voltages and peak voltages, associated with a direct time analysis, and
fundamental voltage component using frequency analysis. No mater what kind of method
was selected to apply, the mechanism of tracking voltage along the signal ,which leads to
time characteristic of magnitude or phase variation in time, was the same – shifting an
executive window. It means that the main algorithm responsible for magnitude estimation
was applied only for a particular portion of samples, designated by the window, many times
for its particular location. Finally, we have achieved many half or one-cycle local values of
rms voltage associated with a position of the window along time axis. Representation of
130
these local rms values along the time axis gives desirable characteristic of voltage variation
during the event.
Until now, we have applied high quality calculation, where the shift of the window was fine
i.e. sample by sample. In a practical adaptation of the rms estimation high quality mode is
too time-consuming and requires computational power. Thus, for monitoring mode, we
often reintroduce the ideas of rms algorithms as the rms value is calculated once per cycle
and can be grouped and averaged in interval of 200ms or 10 minutes [3],[4],[6]. In a
monitoring mode the shift is “fat” i.e. window by window with an optional averaging period
or not . Please notice that this time, the first value of estimated rms voltage is localized in
time moment equals N*dt where dt is the sampling rate. The next points of desirable rms
voltage characteristic are calculated on the basis of fragments of the recorded data,
designated by shifting cycle-by-cycle window. Thus, we need to recalculate the time axis
using “fat” time step N*dt, t=[N*dt, 2*N*dt, …,k*N*dt…]. The description of the idea is
depicted in Fig. 8.16. In comparison with Fig. 8.3, we can observe rough estimation, that
however saves the computational power for calculation.
Additionally, the averaging time is introduced concerning one value for selected numbers of
cycles (half-cycles) i.e. 200ms averaging time means that we achieve one value of rms from
10 one-cycle values or 20 half-cycles values [3],[4],[6].
In order to close the idea, Table 8-6 and Fig. 8.17 show demonstration algorithm or classical
rms calculation working in a monitoring mode without additional averaging. To understand
the difference between high-quality and monitoring mode, we recommend comparing Fig.
8.17 and Fig. 8.5.
131
0 0.05 0.1 0.15 0.2 0.25 0.3
-1
-0.5
0
0.5
1
time[s]
volta
ge [p.u.]
k=1
v(1)v(2)...
V(N)
Vrms(1)
k=1
v(M-N)
...V(M)
Vrms(M/N-1)
In monitoring mode the window is shifted cycle by
cycle
M - total number of samples
N - number of samples take into calculation of local rms value/number of samples of the window
(M/N-1) - number of samples of voltage magnitudecharacteristic in monitoring mode
Vrms(1) Vrms(M/N-1)
0.3398
local application of
the algorithm
v(N)v(N+1)
...V(2N)
Vrms(2)
k=2
0.05 0.1 0.15 0.2 0.25 0.3
0.7
0.75
0.8
0.85
0.9
0.95
1
time [s]
volta
ge m
agnitu
de [p.u.]
Vrms1: one-cycle window (20ms) - MONITORING MODE
0.02TIME STEP=N*dt=N*1/fs
Fig. 8.16. Mechanism of rms voltage characteristic in monitoring mode without averaging
132
Table 8-6. Demonstration algorithm applying classical rms calculation method in monitoring mode without
additional averaging
%---------SIGNAL------------------------------------------------------------- clear all, close all load sag_gd; %sag_gd.mat contain uL1 uL2 uL3 4096 samples %fs=10240Hz %it covers 20 cycles (period of fundamental component) %4096 total number of samples %2048 samples/ 10cycle %204.8 samples/ 1cycle, round 204 sample/cycle x=uL1; h=plot(t,x) set(h,'LineWidth',2); xlabel('time [s]') ylabel('voltage [V]') grid on axis('tight') %---------INITIAL PARAMETERS-------------------------------------------------------------- icycle=1; %icycle- executive window width in cycles of fundamental component fo=50; %fundamental component 50Hz To=1/fo; %one cycle of the fundamental component in [s] tN=icycle*To; %executive window width in [s] N=round(icycle*To/dt) %executive window width in [samples] %----------------------------------------------------------------------- %----------------------------------------------------------------------- %TIME ANALYSIS - Vrms1- rms voltage: 1-phase, monitor mode %----------------------------------------------------------------------- %Vrms1 - vector of local rms voltage %calculated on the basis of N samples of the signal %High quality mode - every next Vrms1 are obtained %by shifting the window along signal sample by sample. %Note - first value of Vrms1 can be obtained after recording of %assumed N samples of executive window (tN in seconds). %Thus we should rescale time axis in Figure representing rms voltage %starting from tN and finish with end of the signal ->t1 %For high quality mode we preserve time step in t1 as dt %For monitoring mode time step equals number of cycles -> window by widnow l=1; %monitoring mode storage initial for k=1:N:length(x)-N k n=[k:k+N-1]; %identification of windowed samples xN=[x(n)]; %windowed signal samples taken for calculation of local rms Vrms1(l)=sqrt(1/N*sum(x(n).^2)); l=l+1; end t1=t(N+1:N:length(x)); figure h=plot(t1,Vrms1,'--bs','MarkerEdgeColor','k','MarkerFaceColor','b','MarkerSize',8) set(h,'LineWidth',2) xlabel('time [s]') ylabel('voltage magnitude [p.u.]') title(['Vrms1: ',num2str(icycle),'-cycle window']) grid on axis('tight') %-----------------------------------------------------------------------
0 0.05 0.1 0.15 0.2 0.25 0.3 0.35
-300
-200
-100
0
100
200
300
time [s]
volta
ge [V]
0.05 0.1 0.15 0.2 0.25 0.3 0.35
100
120
140
160
180
200
220
240
time [s]
volta
ge m
agnitude
[p.u.]
Vrms1: 1-cycle window
Fig. 8.17. Investigated one-phase signal and its rms trend using one-cycle classical rms algorithm in monitoring
mode without additional averaging
133
8.7. Dip transfer in power systems
The prominent number of dips in power systems comes from short-circuit faults which can
appear in a different level of the system structure: transmission system faults, remote
distribution systems faults, local distribution system faults. Other origins of the dips may
have their sources in a starting process of large motors, short interruption and fuses
[1],[2],[3],[4].
According to the standards, dips can be entirely defined by the magnitude and duration.
Considering the duration of dips, we can distinguish self-decaying events and events
determined by fault-clearing equipments. Generally speaking, the faults in transmission
systems are cleared faster than faults in distribution systems. In transmission systems, the
critical fault-clearing time is rather small. Thus, fast protection and fast circuit breaker are
essential. Also transmission and subtransmission systems are normally operated as a grid,
requiring a distance protection or a differential protection, both of which are rather fast. The
principal form of a protection in distribution systems are overcurrent protection. This often
requires some time-grading which increases the fault –clearing time. An exception
constitutes systems in which current-limiting fuses are used that have the ability to clear a
fault within one half-cycle [1],[2],[3],[4].
The drop of rms voltage during the dips depends on a few conditions. One of the crucial
issue is relation between the location of the fault and the location of observation in the
system structure i.e. point of common coupling (PCC). A general rule indicates significant
influence of the dips on the customers grouped at the same level of voltage as appearing
fault. Further distribution of the dip depends on the location of the fault in the system
structure and the presence of the impedance of the transformer. Thus, a short-circuit fault in
local distribution networks (medium or low voltage) typically leads to deep dips whpse injury
every customer grouped in near regions. This is due to small impedance taken into
consideration which comes only from impedance of distribution feeders often with limited
length. These dips will not cause much voltage drop for the customers fed at the sub-
transmission and transmission level. The impedance of the transformers between the
distribution and subtransmission, and further, between the sub-transmission and
transmission level are large enough to considerably limit the voltage drop at the high voltage
side of the transformer. However, the fault in the sub-transmission system will cause a deep
dip in the subtransmission substation and for all customers fed from this substation. The
134
dips will be visible at every lower voltage side of the substation. Customers working on the
transmission voltage level will experience a shallow dip, again due to the transformer
impedance. Finally, the fault in the transmission system will cause a serious dip for fed
substations bordering the faulted line. This dip is then transferred down to all customer fed
from these substations [1],[2],[3],[4].
In order to emphasize mentioned general rules of the dips transfer in the power systems, we
have attached an example of the power network with possible fault position Fig. 8.18 and
the table of scenarios dedicated to dips influence Table 4-1. It must be mentioned that given
characterization has a general nature and does not include some specific situations as
distributed generation.
voltageanalyzer
IndustrialDistributionSystem
Customer B
Equivalentof thesupplysystem
Transmission
Subtransmission
Distribution
Low voltage
Customer A
Customer B
Customer DCustomer C
Workshop nr 3
PCC
Fault 1
Fault 2
Fault 3
Fault 2a
Fault 3a
Fault 5Fault 4
Fig. 8.18. Example of power network with possible fault position scenarios dedicated to general dips transfer
135
Table 8-7 Possible scenarios of dips transfer in power systems from Fig. 8.18 referring to fault position
Label Fig.
8.18 Customer A Customer B Customer C Customer D
Transmission Fault 1 deep dip deep dip deep dip deep dip
Subtransmission Fault 2 not much deep dip deep dip deep dip
Subransmission/
industrial
network
Fault 2a deep dip not much not much not much
Distribution Fault 3 not much not much deep dip deep dip
Distribution/
industrial
network
Fault 3a not much deep dip not much not much
Distribution
utility network Fault 4
rather not
notice
rather not
notice deep dip shallow dip
Distribution
utility network Fault 5
rather not
notice
rather not
notice shallow dip deep dip
As an example, we would like to consider an often met scenario – a fault in a local
distribution network. Range of the investigation is scoped around influence on “neighbour”
customer. It corresponds to scenario Fualt 3a from Fig. 8.18 and Table 4-1, where
measurements are localized at customer, fed by a feeder without the fault. Considering the
mentioned scenario, we can scope on some details as influence of the local parameters on
character of the dips observed by the customer, especially relation between the dips and
length of the faulted feeder as well as relation between the dips and kind of fault. The aim of
the activities is to familiarize with a selected origin of the dips. In order to perform the
investigations we have prepared the model of the simplified diagram of the network
illustrated in Fig. 8.19. Presented here diagram was modelled in Matlab (R13 6.5) Simulink
environment using SimPowerSystem Blockset [7] ,[8]. The model is available at
http://eportal.eny.pwr.wroc.pl/ .
136
voltageanalyzer
IndustrialDistributionSystem
Customer B
Equivalentof thesupplysystem
PCC
ABC
load2
ABC
load1
ABCN
abcn
Three-phaseTransformer
(Two Windings)
A
B
C
Three-Phase Source
A
B
C
ABC
VabcIabc
Three-PhaseV-I Measurement2
A
B
C
A
B
C
Vabc
IabcThree-Phase
V-I Measurement1
A
B
C
ABC
VabcIabc
Three-PhaseV-I Measurement
Scope3Scope2
Scope1
A
B
C
A
B
C
A
B
C
A
B
C
A
B
C
A
B
C
A
B
C
A
B
C
PI Line1/2
A
B
C
3-Phase Fault
Investigated influence:
Variable length of the faulted feeder,
Line 1, i.e. variable fault position
Investigated influence:
Variable kind of fault
Customer Bvoltageanalyzer
Line 1 - faulted
Line 2 - customer
PI Line1/2
PI Line1/2
PI Line1/2
Fig. 8.19. Simplified diagram of the investigated network and its model in Simulink SimPowerSystem Blockset
The system is described by an initial short-circuit apparent power 3GVA and a voltage level
of 110 kV. The transformer is a two-winding (delta-wye-isolated) 110/15-kV distribution
transformer with a nominal power 10MVA. L1 and L2 are typical overhead lines. Both lines
supply the RL loads. The line 2 has a 5km length and fed observed industrial customer. In a
point of numerical reasons, the lines are modeled as two Pi-line components. The short
circuit occurs at the end of the line L1.
The investigation is aimed at:
- the influence of the length of Line 1 (faulted line) on the dips injured Customer B:
suggested range L1=1km, L1=5km,
- the influence of the kind of fault on the dips injured Customer B: suggested range – A-GND
one-phase fault with ground, A-B two-phase fault.
The investigated signals comes from the observation made by Customer B. In model it is
realized by Scope 2. Setup of the Scope considers sample time 2e-4s (0.2ms). For a digital
calculation, it is very important information which allows defining sampling frequency
fs=1/(0.2ms)=5kHz. The time axis is automatically saved in workspace. Selected range of the
observation is (0-0.6s).
137
As an example Fig. 8.20 represents the character of the dip for the same kind of a fault, A-
GND, but for different fault location L1=1km vs L1=5kme. Estimation method utilizes classical
rms definition in high-quality mode.
0 0.1 0.2 0.3 0.4 0.5 0.6-2.5
-2
-1.5
-1
-0.5
0
0.5
1
1.5
2
2.5x 10
4
time [s]
volta
ge [p.u.]
3-phase voltage: A-GND fault,L1=1km
va
vb
vc
0.05 0.1 0.15 0.2 0.25 0.3 0.35 0.4 0.45 0.5 0.55
2000
3000
4000
5000
6000
7000
8000
9000
10000
time [s]
volta
ge m
agnitude
[p.u.]
Vrms1: 1-cycle window, A-GND fault
L=1km
L=5km
0 0.1 0.2 0.3 0.4 0.5 0.6
-2
-1.5
-1
-0.5
0
0.5
1
1.5
2
x 104
time [s]
volta
ge [p.u.]
3-phase voltage: A-GND fault,L1=5km
va
vb
vc
Fig. 8.20. Influence of distance of faulted line on observed dip in grid point in neighbour line
138
Fig. 8.21 depicts the influence of the kind of fault in faulted line on observed dip in grid point
in neighbour line for the same length of faulted line which equals 1km. Estimation method
utilizes classical rms definition in high quality mode.
0 0.1 0.2 0.3 0.4 0.5 0.6-2.5
-2
-1.5
-1
-0.5
0
0.5
1
1.5
2
2.5x 10
4
time [s]
volta
ge [p.u.]
3-phase voltage: A-GND fault,L1=1km
va
vb
vc
0.05 0.1 0.15 0.2 0.25 0.3 0.35 0.4 0.45 0.5 0.55
2000
3000
4000
5000
6000
7000
8000
9000
10000
time [s]
volta
ge m
agnitude
[p.u.]
Vrms1: 1-cycle window, L1=1km
A-GND
A-B
0 0.1 0.2 0.3 0.4 0.5 0.6-2.5
-2
-1.5
-1
-0.5
0
0.5
1
1.5
2
2.5x 10
4
time [s]
volta
ge [p.u.]
3-phase voltage: AB fault,L1=1km
va
vb
vc
Fig. 8.21. Influence of kind of fault in faulted line on observed dip in grid point in neighbour line
8.8. Literature
[1] Bollen M.H.J., Understanding Power Quality Problems. Voltage sags and interruptions,
IEE Press Series on Power Engineering, 2000.
[2] Dugan R. C. et al, Electrical Power Systems Quality, McGraw-Hill, 2004.
[3] Sankaran C., Power Quality, CRC Press, 2002.
[4] Power Quality Application Guide. Chapter 5. Voltage Disturbances. The European Copper
Institute, Leonardo da Vinci European Program on Power Quality Assessment.
[5] Łobos T., Sikorski T., Amaris H., Alonso M., Florez D., Classical and alternative methods to
voltage quality monitoring, Computer Application in Electrical Engineering, Poznan, 2009.
139
[6] IEC 61000-4-30:2008, Electromagnetic compatibility (EMC): Testing and measurement
techniques -Power quality measurement method.
[7] SimPowerSystems User’s Guide. The Math Works Inc., 2006.
[8] P. Janik, Charakterystyka zakłóceń jakości energii elektrycznej za pomocą
zaawansowanych metod analizy sygnałów, in Polish, Ph.D. Thesis, 2005.
140
9. Computational application of power quality assessment: harmonics
The previous chapter was dedicated to dips assessment, as the representative of events. It is
worth emphasising that main division of power quality disturbances includes events as well as
variations. Events are sudden disturbances with a beginning and an ending. To measure the
events, triggering process must be involved, usually using a threshold. Variations are steady-state
or quasi-steady-state disturbances that require continuous measurements.
With reference to more classical power engineering, the measurement of variations is similar to
metering of the energy consumption (i.e., continuous), whereas the measurement of events is
similar to the functioning of a protection relay (i.e., triggered). Neglecting the purpose of the
measurements, every process of quantification starts from devices responsible for voltage and
current measurements. The difference is in the further processing and application of the
measured signals. The results of power quality monitoring are not used for any automatic
intervention in the system. Exceptions are the measurements as part of power quality mitigation
equipment, but such equipment is more appropriately classified as protection or control
equipment. Fig. 9.1 depicts a general scheme of power quality measurements with distinction
between variation and event .
Fig. 9.1. General scheme of power quality measurements: (1) voltage or current in system; (2) sampled and digitized
voltage or current; (3) quantity for further processing [2]
This chapter is dedicated to variations. The process of variations assessment requires calculation
of appropriate characteristics. This may be the rms voltage trend, the frequency trend or the
141
spectrum. Due to a continuous character of measurement some averaging conditions over a
certain interval must be introduced. The calculation of rms-voltage, the frequency or the spectrum
in high-quality mode would generate huge number of data with high requirements for data
storage and computational power. Thus, for variations much more practical approach consider
monitoring-mode with additional parameter called averaging time. For example, if averaging time
equals 200ms i.e. 10 periods of fundamental 50Hz component, we achieve one averaged value of
rms from ten one-cycles values, or one estimated frequency value from 10-cycle, or at least one
spectrum obtained on the basis of 10-cycle [1],[2].
The standard document IEC 61000-4-30 [4] prescribes the following intervals: 10 or 12 cycles, 150
or 180 cycles, 10 min, and 2 h. Some monitors use different window lengths. Some monitors also
give maximum and minimum values obtained during each interval. Some monitors do not take the
average of the characteristic over the whole interval but a sample of the characteristic at regular
intervals, for example, the spectrum obtained from one cycle of the waveform once every 5 min.
Further postprocessing consists of the calculation of representative statistical values (e.g., the
average or the 95 percentile) over longer periods (e.g., one week) and over all monitor locations.
The resulting values are referred to as site indices and system indices, respectively [2].
9.1. Harmonics distortion and Fourier series parameterization
Waveform distortion includes all deviations of the voltage or current waveform from the ideal sine
wave. But variations in magnitude and frequency are not considered as waveform distortion
although a complete distinction between the different types of variations is not possible. A
number of different forms of waveform distortion can be distinguished:
- harmonic,
- interharmonic,
- nonperiodic distortion.
In most studies, only the harmonic distortion is considered. Nonharmonic distortion
(interharmonics and nonperiodic distortion) is much harder to quantify through suitable
parameters and it is regularly neglected. Another reason for neglecting nonharmonic distortion is
that harmonic distortion dominates in most cases. Coming back to the harmonics we can
distinguish odd-harmonics and even-harmonics. The odd-harmonics manifests itself symmetrically
for the positive and negative half of sine wave. Odd-harmonic distortion is typically dominant in
supply voltage and load current. The result of even-harmonic distortion is that positive and
142
negative half-cycles of the signal are no longer symmetrical. Even-harmonic distortion of voltage
or current is normally rather small.
The measurement of harmonic distortion is defined in two IEC standards: IEC 61000-4-7 [5] and
IEC-61000-4-30 [4]. The first one defines the way in which harmonic current distortion shall be
measured when comparing equipment emission with the emission limit. The latter document
defines the way in which voltage quality shall be measured.
According to the IEC standards, the Fourier series shall be obtained over a rectangular window
with a length equal to 10 cycles in a 50-Hz system and equal to 12 cycles in a 60-Hz system.
However we have to realize that the measurement systems may use different sampling rate. Thus,
given 10-cycle interval would contain different number of samples, more for high sampling rate,
and less for low sampling rate. Here it is worth noticing that in pair with condition given by IEC
runs IEEE document IEEE Standard Common Format for Transient Data Exchange (COMTRADE) for
Power Systems [6]. This document gives some hints for sampling rate and number of samples in
one-cycle of basic components, that can be easily recalculate for 10 cycles, Fig. 9.2.
Fig. 9.2. Sampling rates for 50Hz system and 60Hz system with its relations to number of samples in one-cycle of
fundamental component [6]
The assessment of the harmonic and interharmonic distortion is based on the discrete Fourier
spectrum. The output of the discrete Fourier transform consists of a number of frequency
143
components, both harmonics and interharmonics. The efficient assessment of the spectrum
requires some initial calculation and preparation in order to associate the output of Fourier series
with signal parameters. Main activities are scoped around frequency bin specification. An
undermentioned example shows a required initial calculation so that performed Fourier series
might have a physical meaning.
Example: initial calculation for Fourier series parameterization
Given: sampling rate fS=4kHz, system f0=50Hz, interval 10-cycle
Searched: a number of samples in the interval and a total number of frequency bins
N=?,frequency resolution of DFT algorithm dfF=?, frequency axis f=[….], maximum order of
harmonics H=?, frequencies of particular harmonics hf, index of the harmonics at the frequency
axis nfch=[…].
Answer:
Referring to Fig. 9.2 for given sampling rate fS=4kHz we have 80 samples per fundamental
component 50Hz.
The interval includes 10 cycles, thus for a calculation of DFT we take N=10x80=800 samples.
Frequency resolution is then dfF=fS/N=4000/800=5Hz.
According to “Nyquist theory” frequency observation range is limited to 1/2 fS=2kHz. Thus, we can
generate the frequency bins for the frequency axis f=[0:dfF: 1/2 fS]=[0:5:2000] – length of
frequency axis is 401 bins.
Please notice that total number of frequency bins is N=800 and the length of frequency
observation is limited to 401 bins. Remember that performing the DFT, you achieve the
information about positive and negative part of frequency axis. For the real signal the spectrum is
symmetrical and desirable information of the spectrum can be reached from the positive part of
frequency axis – in case of DFT algorithm it is first N/2+1 elements.
Maximum order of harmonics can be calculated as H=(1/2fs)/f0=2000/50=40.
Frequencies of particular harmonics hf=[1:1:H]*f0=[50,100,150….2000] in Hz.
Index of the harmonics in the linear spectrum nfch=([1:1:H]*f0/dfF)+1=[11,21, …. 401]
Finally, frequency bin is characterised which depicts Fig. 9.3.
0 5 10 15 20 25 30 35 40 45 50h=1
55
1 2 3 4 5 6 7 8 9 10 11h=1
12
frequency axis
frequency index/index of harmonics
60 65 70 75 80 85 90 95 100h=221h=2
13 14 15 16 17 18 19 20
196019651970 19751980 19851990 1995
393 394 395 396 397 398 399 400
2000h=40
401h=40
1955
392
f
Fig. 9.3. Example of frequency bins characterisation
144
Some additional comments:
The result of applying the DFT is a spectrum with 5-Hz spacing between frequency components.
The spectrum thus contains both harmonics and interharmonics.
We can distinguish the spectrum estimation for voltage and current. The oltage quality is how the
network affects the customer or the load; the current quality is how the customer or load affects
the network. A similar distinction holds for waveform distortion: the distorted voltages affect the
customer equipment; distorted currents affect the network components.
9.2. Investigated phenomena
Main activities in this chapter are concentrated on the investigation of waveform distortion
caused by very specific non-linear load which is dc-arc funace. An electric arc furnace (EAF) is a
furnace that heats charged material by means of an electric arc. Arc furnaces range in size from
small units of approximately one ton capacity (used in foundries for producing cast iron products)
up to about 400 ton units used for secondary steelmaking. Arc furnaces used in research
laboratories and by dentists may have a capacity of only a few dozen grams. Electric arc furnace
temperatures can be up to 1,800 degrees Celsius. Arc furnaces differ from induction furnaces in
that the charge material is directly exposed to the electric arc, and the current in the furnace
terminals passes through the charged material. A typical dc arc furnace plant and an example of
practical realization is shown in Fig. 9.4.
From the electrical point of view, a furnace plant consists of dc arc furnace connected to a
medium volt-age ac busbar with two parallel thyristor rectifiers that are fed by transformer
secondary winding with Δ and Υ connection, respectively. The medium voltage busbar is
connected to the high voltage busbar with a HV/MV transformer whose wind-ings are Υ-Δ
connected. The power of the furnace is 80 MW. The other parameters are: transformer T1-80
MVA, 220kV/21kV; transformer T2- 87 MVA, 21kV/0.638kV/0.638kV. Some filters are provided in
the plants.
In the dc arc furnace, the presence of the ac/dc static converter and the random motion of the
electric arc, whose non linear and time-varying nature is known, are responsible for dangerous
perturbations, in particular waveform distortions and voltage fluctuations. In particular, the
behaviour of the DC arc furnace can lead to the aperiodicity of AC electrical quantities, namely
voltage and current at MV busbar. The phenomena correlated to the arc behavior are very
complex. In [7],[8] it has been shown that the DC arc voltage waveform has the aperiodic and
145
irregular behaviour that characterizes every chaotic phenomenon. As an example, Fig. 9.5 shows
an example of current waveform and voltage waveform at MV busbar.
a)
T1 T2
HV
L
-
ia
ia,ref+regulator
firingcircuit
MV
measurement
b)
Fig. 9.4. Dc-arc furnace plant (a) and an example of practical realization of arc furnace (b)
0 0.02 0.04 0.06 0.08 0.1 0.12 0.14 0.16 0.18-2500
-2000
-1500
-1000
-500
0
500
1000
1500
2000
2500
time [s]
i [A]
current in dc-arc furnace
0 0.02 0.04 0.06 0.08 0.1 0.12 0.14 0.16 0.18
-1.5
-1
-0.5
0
0.5
1
1.5
x 104
time [s]
v [V
]
voltage in dc-arc furnace
Fig. 9.5. Fragments of current and voltages at MV busbar of dc-arc furnace plant [7],[8]
10-cycle interval of one-phase current (i) and voltage (u) at MV busbar, with sampling rate equals
5kHz is stored in DCArcFurnacemeasure0_02.mat. This file contains also some additional utilizing
parameters as: time vector (t), sampling rate fS=5kHz, and time step (time resolution) (dt=1/fS).
The file is available at http://eportal.eny.pwr.wroc.pl/
146
9.3. “Linear” spectrum estimation using FFT:
Spectrum estimation is considered as a digital representative of continuous “linear” spectrum,
which includes whole spectrum components, including harmonics and interharmonics. Table 9-1
consists of demonstration algorithm applying FFT in Matlab in order to calculate the spectrum of
the voltage in example of investigated dc-arc furnace plant.
Table 9-1. Demonstration algorithm applying FFT in order to reveal “linear” spectrum
%---------SIGNAL------------------------------------------------------------- clear all, close all cd .. cd Data load DCArcFurnacemeasure0_02.mat; cd .. cd Programs %file contains 10-cycles 50Hz of dc-arc furnace measured signals: %sampling rate fs=5kHz; %Thus we have 100 samples per cycle % and 1000 total number of samples in 10 cycles %i - current, %u - voltage, %Additionally: %fs=5000; [Hz] - sampling rate %dt=1/fs; [s] - sampling interval %t-time vector [0:dt:0.2-dt]; x=u; disp('-----------------------ANALYSED SIGNAL---------------------------------------------------------') disp(' ') disp('CONFIRMATION OF SAMPLED DATA:') disp(['Sampling frequency fs=',num2str(fs),' Hz']) disp(['Time step dt=',num2str(dt),' s']) disp(['End of signal in samples:',num2str(length(x)),' ']) disp(['End of signal in seconds:',num2str(length(x)*dt),' s']) figure h=plot(t,x); set(h,'LineWidth',2); xlabel('time [s]') ylabel('u [V]') title('voltage in dc-arc furnace') grid on axis('tight') %--------------------------------------------------------------------- %---------INITIAL PARAMETERS-------------------------------------------------------------- f0=50; %[Hz]; %fundamental component N=length(x); %number of frequency bins dfF=fs/(N); %frequency resolution %----------------------------------------------------------------------- %----------------------------------------------------------------------- %FREQUENCY ANALYSIS %----------------------------------------------------------------------- X=fft(x,N); %Fourier spectrum (positive and negative frequencies) X_amp=abs(X)*(2/N); %magnitude spcectrum with scaling factor X_amp_positive=X_amp(1:1:N/2+1); %magnitude spcectrum only for positive f f_positive=dfF*(0:1:N/2); %positive f %full range of frequency observation 0-fs/2Hz fc1=0; fc2=fs/2; nfc1=round(fc1/dfF)+1; nfc2=round(fc2/dfF)+1; figure h=plot(f_positive(nfc1:nfc2),X_amp_positive(nfc1:nfc2)); set(h,'LineWidth',2); xlabel('f[Hz]') ylabel('magnitude'); %title('spectrum of current in dc-arc furnace') title('spectrum of voltage in dc-arc furnace') grid on; %zoom in frequency 0-800Hz fc1=0; fc2=800; nfc1=round(fc1/dfF)+1; nfc2=round(fc2/dfF)+1; %zoom in magnitud 0-400 Ampc1=0; Ampc2=400; figure h=plot(f_positive(nfc1:nfc2),X_amp_positive(nfc1:nfc2)); set(h,'LineWidth',2); xlabel('f[Hz]') ylabel('magnitude'); %title('spectrum of current in dc-arc furnace - zoom') title('spectrum of voltage in dc-arc furnace - zoom') axis([fc1, fc2, Ampc1, Ampc2]) grid on; %-----------------------------------------------------------------------
147
After calculation of FFT we obtain a set of the frequency component concentrated on every 5Hz
frequency bin including its magnitude and phase. Fig. 9.6 and Fig. 9.7 shows magnitude spectrum
of voltage and current, respectively. The next activities in assessment of the frequency component
is separation of harmonics and interharmonics.
0 0.02 0.04 0.06 0.08 0.1 0.12 0.14 0.16 0.18
-1.5
-1
-0.5
0
0.5
1
1.5
x 104
time [s]
v [V
]
voltage in dc-arc furnace
0 500 1000 1500 2000 25000
2000
4000
6000
8000
10000
12000
14000
16000
18000
f[Hz]
mag
nitude
spectrum of voltage in dc-arc furnace
0 100 200 300 400 500 600 700 8000
50
100
150
200
250
300
350
400
f[Hz]
mag
nitude
spectrum of voltage in dc-arc furnace - zoom
zoom:(0-400)V – (0-800)Hz
Fig. 9.6. Investigated one-phase voltage signal in dc-arc furnace plant and its “linear” spectrum using FFT
148
0 0.02 0.04 0.06 0.08 0.1 0.12 0.14 0.16 0.18-2500
-2000
-1500
-1000
-500
0
500
1000
1500
2000
2500
time [s]
i [A]
current in dc-arc furnace
0 500 1000 1500 2000 25000
500
1000
1500
2000
2500
f[Hz]
mag
nitude
spectrum of current in dc-arc furnace
0 100 200 300 400 500 600 700 800 900 10000
10
20
30
40
50
60
70
80
90
100
f[Hz]
mag
nitude
spectrum of current in dc-arc furnace - zoom
zoom:(0-100)A – (0-800)Hz
Fig. 9.7. Investigated one-phase current signal in dc-arc furnace plant and its “linear” spectrum using FFT
149
9.4. Harmonics and subdivision into even and odd harmonics
The output of the discrete Fourier transform consists of a number of frequency components, both
harmonics and interharmonics. However, the harmonics are only the integral multiplication of
fundamental component: 1-st harmonic is 1*50Hz, 2-nd harmonic is 2*50Hz, 3-rd harmonic is
3*50Hz etc. Additionally, the harmonics can be subdivided into even and odd harmonics. Even-
harmonic distortion of voltage or current is normally rather small. In many cases odd-harmonics
have significant meaning.
Literature provides an additional index in order to depict and summarize a harmonics distortion
effect [1],[2],[3],[4],[5]. It is called total harmonic distortion index THD define as:
2
2
1
*100%
H
h
h
C
THDC
==
∑ (9.1)
where Ch – magnitude of h harmonic (can be voltage or current), C1 – magnitude of fundamental
(first harmonic)
The harmonic spectrum can be further subdivided into even harmonics and odd harmonics with
their corresponding THD definitions. The total even-harmonic distortion THDeven may be defined
as:
2*
2
2
1
*100%
h
H
h
C
THDevenC
==
∑ (9.2)
where C2h – magnitude of even harmonic, C1 – magnitude of fundamental (first harmonic)
The total odd-harmonic distortion THDodd may be defined as:
2* 1
2
2
1
*100%
h
H
h
C
THoddC
−==
∑ (9.3)
where C2h-1 – magnitude of odd harmonic, C1 – magnitude of fundamental (first harmonic)
The confirmation of the proper calculation can use below relations:
2 2 2 ; THD THDodd THDeven THD THDodd= + ≈ (9.4)
150
Activities: Perform initial calculation for Fourier series parameterization
Given: sampling rate fS=5kHz, system f0=50Hz, interval 10-cycle
Aim: According to the previous example of the initial calculation for a number of samples in the
interval and total number of frequency bins N=?,frequency resolution of DFT algorithm dfF=?, the
frequency axis f=[….], maximum order of harmonics H=?, frequencies of particular harmonics hf,
index of the harmonics at the frequency axis nfch=[…].
Answer: An expected answer is included in the figure below, Fig. 9.8
0 5 10 15 20 25 30 35 40 45 50h=1
55
1 2 3 4 5 6 7 8 9 10 11h=1
12
frequency axis
frequency index/index of harmonics
60 65 70 75 80 85 90 95 100h=221h=2
13 14 15 16 17 18 19 20
24602465 2470 2475 2480 24852490 2495
493 494 495 496 497 498 499 500
2500h=50
501h=50
2455
492
f
Fig. 9.8. Frequency bins characterization for calculation of investigated voltage and current in example of dc-arc
A definition of frequency bin in “linear” spectrum allows to reveal harmonics, even harmonics and
odd harmonics. In the presented example, harmonics [50,100,150,…2500]Hz are situated at
multiplication of index 11 i.e. [11,21,31 …..501]. Even harmonics is associated with bins with
number [21,41,….501]. Odd harmonics, neglecting a fundamental component, which is localized at
index 11, start from index 31 i.e. [31, 51,71…491]. According to the subdivision, we can represent
now the full harmonic spectrum as well as the harmonics spectrum including even as well as odd
components. The results concerning investigation of harmonics in voltage and current in dc-arc
furnace plant is presented in Fig. 9.9, Fig. 9.10. The calculation of THD indices confirms
predominant contribution of odd harmonics.
151
a) Harmonics relative to
fundamental
b) Even harmonics relative to
fundamental
c) Odd harmonics relative to
fundamental zoom – without fundamental
5 10 15 20 25 30 35 40 45 500
0.5
1
1.5
harmonic order without first harmonic
magn
itude
- %
of fund
amen
tal c
ompo
nen
t
harmonics in voltage in dc-arc furnace - zoom: THD=2.3254%
THD = 2.3254
zoom – without fundamental:
0 10 20 30 40 50 600
0.05
0.1
0.15
0.2
0.25
0.3
0.35
0.4
EVEN harmonic order without first harmonic
magn
itude
- %
of fund
amen
tal c
ompo
nen
t
EVEN harmonics in voltage in dc-arc furnace - zoom: THDeven=0.68978%
THDeven =
0.6898
zoom – without fundamental:
5 10 15 20 25 30 35 40 450
0.5
1
1.5
ODD harmonic order without first harmonic
magn
itude
- %
of fund
amen
tal c
ompo
nen
t
ODD harmonics in voltage in dc-arc furnace - zoom: THDodd=2.2208%
THDodd =
2.2208
Checking: 2 2 2 ; THD THDodd THDeven THD THDodd= + ≈ 2 2 2= + ≈2.3254 2.2208 0.6898 ; 2.3254 2.2208
Fig. 9.9. Harmonics analysis of voltage in dc-arc furnace plant: presentation of harmonics, THD indices calculation
a) Harmonics relative to
fundamental
b) Even harmonics relative to
fundamental
c) Odd harmonics relative to
fundamental zoom – without fundamental
5 10 15 20 25 30 35 40 45 500
0.5
1
1.5
2
2.5
3
3.5
harmonic order without first harmonic
mag
nitude
- %
of fund
amen
tal c
ompo
nent
harmonics in current in dc-arc furnace - zoom: THD=4.9239%
THD: 4.9239
zoom – without fundamental:
5 10 15 20 25 30 35 40 45 500
0.1
0.2
0.3
0.4
0.5
EVEN harmonic order without first harmonic
mag
nitude
- %
of fund
amen
tal c
ompo
nent
EVEN harmonics in current in dc-arc furnace - zoom: THDeven=1.0385%
THDeven =
1.0385
zoom – without fundamental:
5 10 15 20 25 30 35 40 450
0.5
1
1.5
2
2.5
3
3.5
ODD harmonic order without first harmonic
mag
nitude
- %
of fund
amen
tal c
ompo
nent
ODD harmonics in current in dc-arc furnace - zoom: THDodd=4.8132%
THDodd =
4.8132
Checking: 2 2 2 ; THD THDodd THDeven THD THDodd= + ≈ 2 2 2= + ≈4.9239 4.8132 1.0385 ; 4.9239 4.8132
Fig. 9.10. Harmonics analysis of current in dc-arc furnace plant: presentation of harmonics, THD indices calculation
9.5. Groups of harmonics, interharmonic and suharmonic
To deal with the new characteristics of the frequency spectra in voltage and current waveforms,
the International Electrotechnical Commission (IEC) in the 2nd edition of its standard 61000-4-7 [5]
has proposed the use of harmonic groups instead of the individual harmonic frequencies in order
152
to obtain a better estimation of harmonic distortion in modern power supply systems. The
harmonic groups are defined as the grouping of the spectral bins obtained applying the Discrete
Fourier Transform to the samples of voltage or current waveforms taken in rectangular sampling
windows of 10 periods of the fundamental frequency in a 50-Hz system (12 periods in 60-Hz
systems) and using synchronous sampling [2],[3],[4],[5].
Harmonic Group Chg : in a 50-Hz system there are 9 frequency components between two harmonic
orders h and h + 1. The lowest four are added to group h, the highest four to group h +1, and the
middle one is divided equally between the two groups. The harmonic group Chg is defined by
formula:
4
2 2 2 2( ) 5 ( ) ( ) 5
4
1 1
2 2
i
hg nfch h nfch h k nfch h
i
C C C C
=
− − +
=−
= + +∑ (9.5)
where nfch(h) means an index of harmonic order h in linear spectrum.
Hint: according to our previous initial calculation, nfch includes indices associated with particular
harmonics in linear spectrum. e.g. nfch(1)=11 nfch(2)=21 …
Interharmonic Group Cihg : All frequency components between harmonic orders h and h+1 are
combined in an interharmonic group:
9
2 2( )
1
i
ihg nfch h k
i
C C
=
+
=
=∑ (9.6)
A specified interharmonic group is associated with frequencies lower than fundamental. Is is called
subharmonic group Csub1g
1
2 21 (1)
9
i
sub g nfch k
i
C C
=−
+
=−
=∑ (9.7)
Using the previous example of Fourier series parameterization for sampling rate 4kHz and
frequency 5-Hz resolution, we can introduce the definition of the harmonic group order h and
interharmonic group as well as a subharmonic group as it was illustrated in Fig. 9.11.
153
0 5 10 15 20 25 30 35 40 45 50h=1
55
1 2 3 4 5 6 7 8 9 10 11h=1
12
frequency axis [Hz]
frequency index/index of harmonics
60 65 70 75 80 85 90 95 100h=221h=2
13 14 15 16 17 18 19 20
19601965 1970 1975 1980 19851990 1995
393 394 395 396 397 398 399 400
2000H=40
401h=40
1955
392
f
C1g
Sampling rate: 4000Hz; number of possible harmonics H=40
C2g C(H-1)g
0 5 10 15 20 25 30 35 40 45 50h=1
55
1 2 3 4 5 6 7 8 9 10 11h=1
12
frequency axis [Hz]
frequency index/index of harmonics
60 65 70 75 80 85 90 95 100h=221h=2
13 14 15 16 17 18 19 20
196019651970 1975 1980 19851990 1995
393 394 395 396 397 398 399 400
2000H=40
401h=40
1955
392
f
C(1-2)gC(2-3)g C(H-1-H)gCsub1g
Harmonics groupsC hg
Interharmonics groupsC ihg
Fig. 9.11. Illustration of harmonic, interharmonic and subharmonic group definition in 50Hz system with sampling rate
fs=4kHz
In order to practice an idea of grouping the harmonics, we now encourage to define a harmonic,
interharmonic and subharmonic group in case of investigated voltage in current in the dc-arc
furnace plant.
Activities: Perform initial calculation to reveal frequency bins to grouping process of the
harmonics.
Given: sampling rate fS=5kHz, system f0=50Hz, interval 10-cycle
Aim: Selection of frequency bins in order to create a group of harmonics
Answer: An expected answer is included in the figure below, Fig. 9.12.
Fig. 9.12. Harmonic, interharmonic and subharmonic group definition of investigated voltage and current in example
of dc-arc
The results concerning investigation of groups of harmonics in voltage in the dc-arc furnace plant
is presented in Fig. 9.13. If we compare the assessment of the distortion in voltage based only on
the harmonics, Fig. 9.9, we can reveal smaller values of particular components in the process of
the assessment in comparison with its grouping counterpart. Discussed here an approach can have
154
a crucial meaning especially in presence of high value of interharmonics which in classical
calculation, based strictly on harmonics, in neglected.
Fig. 9.14 represents the contribution of particular harmonics and interharmonics groups.
Predominant values are concentrated around the harmonics, however, total neglecting of
interharmonics, as in case of classical harmonics assessment, has a right to stay under the
discussion.
a) Harmonic groups relative to
fundamental
b) Interharmonic groups relative
to fundamental
c) Subharmonic group relative
to fundamental zoom – without fundamental
5 10 15 20 25 30 35 40 450
0.2
0.4
0.6
0.8
1
1.2
1.4
1.6
1.8
2
harmonic group order without first harmonic
magn
itude
- %
of fund
amen
tal c
ompo
nen
t
HARMONIC GROUP in voltage in dc-arc furnace - zoom
zoom – without fundamental:
5 10 15 20 25 30 35 40 450
0.2
0.4
0.6
0.8
1
1.2
1.4
1.6
1.8
interharmonic group order
magn
itude
- %
of fund
amen
tal c
ompo
nen
t
INTERHARMNONIC GROUP in voltage in dc-arc furnace
zoom – without fundamental:
5 10 15 20 25 30 35 40 45 500
0.2
0.4
0.6
0.8
1
harmonic and interharmonic group order
magn
itude
- %
of funda
mental c
ompo
nent
SUBRHARMNONIC GROUP in voltage in dc-arc furnace
Fig. 9.13. Group of harmonics, interharmonics and subharmonic of voltage in dc-arc furnace plant
5 10 15 20 25 30 35 40 45 500
10
20
30
40
50
60
70
80
90
100
harmonic and interharmonic group order
mag
nitude
- %
of funda
mental compo
nen
t
SUBRHARMNONIC,HARMONIC,INTERHARMONIC GROUP in voltage in dc-arc furnace
subharmonic group
harmonic grops
interharmonic groups
Fig. 9.14. Contribution of particular groups of harmonics, interharmonics and subharmonic of voltage in dc-arc
furnace plant
155
Mentioned above discussion under contribution of interharmonics in assessment of the
harmonic distortion has a more significant meaning in case of current, which can be much
more affected by the harmonics than voltage. The level of harmonics defined as a group and
as selected one frequency bin can be recognized comparing Fig. 9.15 and Fig. 9.10.
Additionally, contribution of particular harmonics and interharmonics groups are presented
in Fig. 9.16.
a) Harmonic groups relative to
fundamental
b) Interharmonic groups relative
to fundamental
c) Subharmonic group relative
to fundamental zoom – without fundamental
5 10 15 20 25 30 35 40 450
1
2
3
4
5
harmonic group order without first harmonic
magn
itude
- %
of fund
amen
tal c
ompo
nen
t
HARMONIC GROUP in current in dc-arc furnace - zoom
zoom – without fundamental:
5 10 15 20 25 30 35 40 450
0.5
1
1.5
2
2.5
3
3.5
4
4.5
5
interharmonic group order
magn
itude
- %
of fund
amen
tal c
ompo
nen
t
INTERHARMNONIC GROUP in current in dc-arc furnace
zoom – without fundamental:
5 10 15 20 25 30 35 40 45 500
0.5
1
1.5
2
2.5
3
3.5
4
4.5
harmonic and interharmonic group order
mag
nitude
- %
of fund
amen
tal c
ompo
nent
SUBRHARMNONIC GROUP in current in dc-arc furnace
Fig. 9.15. Group of harmonics, interharmonics and subharmonic of current in dc-arc furnace plant
5 10 15 20 25 30 35 40 45 500
10
20
30
40
50
60
70
80
90
1 00
harmonic and int erharmonic group order
mag
nitude
- %
of fundam
ental c
ompo
nent
SUBRHARMNONIC,HARMONIC,INTERHARMONIC GROUP in curr ent in dc- arc furna ce
subharmonic gro up
harmonic gropsinterha rmonic gr oups
Fig. 9.16. Contribution of particular groups of harmonics, interharmonics and subharmonic of current in dc-arc
furnace plant
156
9.6. Literature
[1] Bollen M.H.J., Understanding Power Quality Problems. Voltage dips and interruptions,
IEE Press Series on Power Engineering, 2000.
[2] Bollen M.H.J, Irene Yu-Hua Gu, Signal Processing of Power Quality Disturbances, 2006
The Institute of Electronics and Electrical Engineers, Inc.
[3] Power Quality Application Guide. Chapter 3. Harmonics. The European Copper Institute,
Leonardo da Vinci European Program on Power Quality Assessment.
[4] Electromagnetic compatibility (EMC), Part 4, Section 30: Power quality measurement
methods, IEC 61000-4-30.
[5] Electromagnetic compatibility (EMC), Part 4, Section 7: General guide on harmonics and
interharmonics measurements and instrumentation, for power supply systems and
equipment connected thereto, IEC 61000-4-7.
[6] IEEE C37.111-1991 - IEEE Standard Common Format for Transient Data Exchange
(COMTRADE) for Power Systems
[7] Bracale A., Carpinelli G., Lauria D., Leonowicz T., Lobos T., Rezmer J., On Some Spectrum
Estimation Methods for Analysis of Non-Stationary Signals in Power Systems. Part I:
Theoretical Aspects, Part II – Numerical applications, 11-th International Conference on
Harmonics and Quality of Power, Lake Placid, New York 2004, Session: Harmonic
Propagation, paper no. - “f” and “g.
[8] Łobos Tadeusz, Sikorski Tomasz, Schegner Peter: Joint time-frequency representation of
non-stationary signals in electrical power engineering. 15th Power Systems Computation
Conference - PSCC. CD-ROM Proceedings, 2005.
157
10. Elements of power quality report
The construction of a power quality report contains the assessment of set of many
parameters discussed in this book. Recorded values are compared with recommended
values. Standard contents of power quality report includes assessment of :
power frequency
-slow voltage variation
fast voltage variation or flicker severity
unbalance
harmonics
recorded power quality events and transients
Methodology of preparing the data base for assessment of power quality is based usually on
one week assessment period. However requirements for aggregation rules, including
averaging intervals, for different parameters are different [1],[2],[3].
An Algorithm for frequency estimation is based on ten-cycle series. In 50Hz systems, for
every 200ms new value of frequency is stored in monitoring mode. However for assessment
aggregation of 10s averaging interval is introduced, giving set of 60480 per week.
Set of frequency values in one week based on 10s aggregation:
10 1min 1 1 11 *6 *60 *24 *7 60480
s h day week=
In case of rms calculation monitoring the algorithm is based on half or one-cycle series. In
the next step, 10 cycle intervals are aggregated and next averaged referring 10 min interval.
The trend of voltage variation is finally based on a set of 10 min values that gives 1008 values
per week:
Set of rms in one week based on 10 min aggregation:
10min 1 1 11 *6 *24 *7 1008
h day week=
The calculation of unbalance coefficient and harmonics use 10-cycle interval according to
requirements for Fourier series. For the assessment of 10-cycle values are averaged in 10
min as in case of rms trend. Thus, similar to slow voltage variation, a set of unbalance and
harmonics is based on 1008 values per week.
158
Set of unbalance and harmonics in one week based on 10 min aggregation:
10min 1 1 11 *6 *24 *7 1008
h day week=
Short-term flicker coefficient is stored with 10 min aggregation interval. Evaluation of
voltage mitigation is defined by long-term flicker coefficient which is based on two hours
averaging of 10 min short-term flicker coefficient. Thus, evaluation of flicker severity uses 84
values of long-term flicker coefficient for one-week assessment period.
Set of long-term flicker coefficient in one week based on 2h aggregation:
2 1 11 *12 *7 84h day week =
The mentioned rules of aggregations allows to prepare set of values in the assessment
period and perform the evaluation with comparison to referring values [3].
The aim of this chapter is to present an example of power quality report including the
assessment of mentioned crucial parameters. Measurements were done in the main point of
the supply of the factory which has an electronic assembly line including robots and power
electronics. Localisation of the power quality recorder in the distribution station of the
factory presents Fig. 10.1.
15/0.4kV
N
PE
PQ Analyzeri
u
UPSGPS
ia,ib,ic
ua,ub,uc
Reactive powercompensation
3
Swithg
ear1
Swithg
ear
2
Swithg
ear
3
Swithg
ear
3
Fig. 10.1. Localisation of the power quality measurements in distribution station
159
10.1. Power frequency assessment
f mean
49,9
49,92
49,94
49,96
49,98
50
50,02
50,04
50,06
50,08
50,1
So10-03-0600:00
N10-03-0700:00
Pn10-03-0800:00
Wt10-03-0900:00
Śr10-03-1000:00
Cz10-03-1100:00
Pt10-03-1200:00
So10-03-1300:00
[Sb 2009-03-06 - Pt 2009-03-12]
[Hz]
Fig. 10.2. Visualisation of power frequency variation
Table 10-1. Assessment of power frequency variation
Frequency assessment f
[Hz]
f10s min 49,886
f10s Percentile 99.5% 50.055
f10s max 50,131
Visualization of the recorded set of frequency values is shown in Fig. 10.2. The standard EN
50160 indicates that the nominal frequency of the supply voltage shall be 50 Hz. Under
normal operating conditions, the mean value of the fundamental frequency measured over
10 s shall be within a range of
– for systems with synchronous connection to an interconnected system:
50 Hz ± 1 % (i.e. 49,5 Hz... 50,5 Hz) during 99,5 % of a year;
50 Hz + 4 % / - 6 % (i.e. 47 Hz... 52 Hz) during 100 % of the time;
Table 10-1 contains some statistical parameters which have confirmed that in the
investigated case, requirements for frequency standards are fulfilled. During 100% of the
assessment, period frequency values hold in range from 49,886 to 50,131 Hz.
160
10.2. Supply voltage variation
V RMS [V]
233
234
235
236
237
238
239
240
So10-03-0600:00
N10-03-0700:00
Pn10-03-0800:00
Wt10-03-0900:00
Śr10-03-1000:00
Cz10-03-1100:00
Pt10-03-1200:00
So10-03-1300:00
[Sb 2009-03-06 - Pt 2009-03-12]
V
VL1
VL2
VL3
Fig. 10.3. Visualisation of supply voltage variation
Table 10-2. Assessment of supply voltage variation
Voltage assesment VL1 VL2 VL3
[V] [V] [V]
U10min
min 233,94 233,43 234,55
Percentile 95% 237,87 237,51 238,73
max 238,83 238,24 239,40
Visualization of the recorded set of supply voltage values is shown in Fig. 10.3. The standard
EN 50160 indicates that under normal operating conditions:
- during each period of one week 95 % of the 10 min mean r.m.s. values of the supply
voltage shall be within the range of Un ± 10 %,
- all 10 min mean r.m.s. values of the supply voltage shall be within the range of Un + 10 % /
- 15 %.
Table 10-2 contains some statistical parameters which have confirmed that in the
investigated case requirements for rms supply voltage standards are fulfilled. 95% of the
recorded rms values hold in range from 233,43V to 237,87V. Moreover, 100% of the
recorded values met standard expectation.
161
10.3. Flicker severity
Plt [-]
0
0,2
0,4
0,6
0,8
1
1,2
So10-03-0600:00
N10-03-0700:00
Pn10-03-0800:00
Wt10-03-0900:00
Śr10-03-1000:00
Cz10-03-1100:00
Pt10-03-1200:00
So10-03-1300:00
[Sb 2009-03-06 - Pt 2009-03-12]
[-]
PltL1
PltL2
PLtL3
Fig. 10.4. Visualisation of long-term flicker severity
Table 10-3. Assessment of long-term flicker severity
Flicker assessment L1 L2 L3
[-] [-] [-]
Plt2h
min 0,33 0,32 0,32
Percentile 95% 0,93 0,81 0,86
max 0,99 0,83 0,88
Visualization of the recorded set of long-term flicker coefficient values is shown in Fig. 10.4.
The standard EN 50160 indicates that under normal operating conditions, in any period of
one week the long term flicker severity caused by voltage fluctuation should be Plt ≤ 1 for 95
% of the time.
Table 10-3 contains some statistical parameters which have confirmed that in the
investigated case requirements for flicker severity are fulfilled. 95% of the recorded long-
term flicker values hold in range from 0,32 to 0,93. Moreover, 100% of the recorded values
met standard expectation.
162
10.4. Voltage unbalance assessment
V RMS Unbal, Neg / Pos [%]
0
0,05
0,1
0,15
0,2
0,25
0,3
0,35
0,4
0,45
0,5
So10-03-0600:00
N10-03-0700:00
Pn10-03-0800:00
Wt10-03-0900:00
Śr10-03-1000:00
Cz10-03-1100:00
Pt10-03-1200:00
So10-03-1300:00
[Sb 2009-03-06 - Pt 2009-03-12]
%
Fig. 10.5. Visualisation of unbalance variation
Table 10-4. Assessment of unbalance variation
Unbalance assesment „Neg/Pos”
[%]
„Neg/Pos”10min
min 0,15
Percentile95% 0,40
max 0,47
Visualization of the recorded set of unbalance values is shown in Fig. 10.5. The standard EN
50160 indicates that Under normal operating conditions, during each period of one week, 95
% of the 10 min mean r.m.s. values of the negative phase sequence component
(fundamental) of the supply voltage shall be within the range 0 % to 2 % of the positive
phase sequence component (fundamental). In some areas with a partly single phase or two
phase connected network users' installations, unbalances up to about 3 % at three-phase
supply terminals occur.
Table 10-4 contains some statistical parameters which have confirmed that in the
investigated case requirements for unbalance standards are fulfilled. 95% of the recorded
unbalance coefficient hold in range from 0,15 to 0,40. Moreover, 100% of the recorded met
standard expectation.
163
10.5. Harmonics assessment
Table 10-5. Assessment of harmonic variations
Harmonic assessment
Percentile95% Max
Order Limits L1 L2 L3 L1 L2 L3
Nr. [ % ] [ % ] [ % ] [ % ] [ % ] [ % ] [ % ]
2 0.00 - 2.00 0.12 0.21 0.17 0.13 0.22 0.17
3 0.00 - 5.00 0.36 0.40 0.35 0.38 0.44 0.38
4 0.00 - 1.00 0.26 0.36 0.24 0.27 0.39 0.26
5 0.00 - 6.00 1.07 1.55 1.03 1.21 1.73 1.24
6 0.00 - 0.50 0.31 0.52 0.35 0.34 0.57 0.38
7 0.00 - 5.00 1.30 1.67 1.46 1.69 2.05 1.80
8 0.00 - 0.50 0.37 0.49 0.62 0.46 0.54 0.70
9 0.00 - 1.50 0.49 0.66 0.79 0.69 0.76 0.96
10 0.00 - 0.50 0.44 0.42 0.56 0.54 0.45 0.63
11 0.00 - 3.50 1.46 1.23 1.45 2.02 1.68 1.91
12 0.00 - 0.50 0.61 0.36 0.51 0.76 0.51 0.67
13 0.00 - 3.00 0.85 0.73 0.63 1.02 0.83 0.68
14 0.00 - 0.50 0.56 0.62 0.53 0.63 0.73 0.58
15 0.00 - 0.50 0.43 0.53 0.43 0.49 0.69 0.55
16 0.00 - 0.50 0.49 0.45 0.35 0.67 0.78 0.59
17 0.00 - 2.00 0.51 0.38 0.43 0.79 0.69 0.59
18 0.00 - 0.50 0.36 0.20 0.33 0.66 0.38 0.50
19 0.00 - 1.50 0.27 0.24 0.31 0.54 0.39 0.65
20 0.00 - 0.50 0.22 0.12 0.23 0.52 0.23 0.55
21 0.00 - 0.50 0.22 0.12 0.17 0.45 0.29 0.42
22 0.00 - 0.50 0.17 0.10 0.12 0.37 0.22 0.27
23 0.00 - 1.50 0.14 0.07 0.11 0.30 0.17 0.21
24 0.00 - 0.50 0.10 0.06 0.07 0.20 0.11 0.12
25 0.00 - 1.50 0.09 0.07 0.06 0.18 0.12 0.09
164
Harmonics V RMS rel to h01 of h [%]
0
0,2
0,4
0,6
0,8
1
1,2
1,4
2 4 6 8 10 12 14 16 18 20 22 24 26 28 30 32 34 36 38 40 42 44 46 48 50
Order
%
L1
L2
L3
Fig. 10.6. Visualisation of harmonic spectrum averaged in one week
Table 10-5 contains limits for particular harmonics defined in EN 50160 and some statistical
parameters of recorded data. Distribution of the harmonics is also visible in Fig. 10.6.
Classical for LV supply network contribution of odd harmonics, especially 3-rd, 5-th, 7-th, 11-
th is confirmed. There are dominated components. Performed measurements have detected
values over the limits in case of even harmonics as 6-th, 8-th, 10-th,12-th,14-th. The
conclusion concerning harmonics assessment is that the requirements for harmonics
contribution is not fulfilled in the investigated case. Further tests shown that exceeded
values appear temporary and come from resonance effects in the supply in companion with
capacitor banks under process of reactive power compensation.
165
THD V rel to h01 [%]
0
0,5
1
1,5
2
2,5
3
3,5
So10-03-0600:00
N10-03-0700:00
Pn10-03-0800:00
Wt10-03-0900:00
Śr10-03-1000:00
Cz10-03-1100:00
Pt10-03-1200:00
So10-03-1300:00
[Sb 2009-03-06 - Pt 2009-03-12]
%
L1
L2
L3
Fig. 10.7. Visualisation of THDV variation
Table 10-6. Assessment of THDV variation
THDV assesment THDVL1 THDVL2 THDVL3
[%] [%] [%]
THDV
10min
min 1,16 1,34 1,14
Percentile95% 2,36 2,68 2,59
max 2,95 3,18 2,78
Visualization of the recorded set of total harmonics distortion in voltage is shown inFig. 10.7.
The standard EN 50160 indicates THD of the supply voltage (including all harmonics up to
the order 40) shall be less than or equal to 8 %.
Table 10-6 contains some statistical parameters of THDV which have confirmed that in the
investigated case requirements are fulfilled. 95% of the recorded THDV coefficient hold in
range from 1,14 to 2,68. Moreover, 100% of the recorded values met standard expectation.
166
10.6. Summary of the report
Presented in the previous section, the assessment of crucial power quality parameters
indicates that:
in case of power frequency, requirements are fulfilled
in case of voltage supply, requirements are fulfilled
in case of flicker severity, requirements are fulfilled
in case of voltage unbalance, requirements are fulfilled
in case of THD of the supply voltage, requirements are fulfilled
in case of harmonics requirements are not fulfield
Performed measurements have detected values over the limits in case of even harmonics as
6-th, 8-th, 10-th,12-th,14-th. It is recommended to make the test of the cooperation the
capacitor banks with the supply during the regulation process of reactive power
consumption.
During the assessment period no voltage events were detected.
10.7. Literature
[1] IEC 61000-4-30:2008, Electromagnetic compatibility (EMC): Testing and measurement
techniques -Power quality measurement method.
[2] Bollen M.H.J, Irene Yu-Hua Gu, Signal Processing of Power Quality Disturbances, 2006
The Institute of Electronics and Electrical Engineers, Inc
[3] EN 50160: Voltage characteristics of electricity supplied by public distribution networks
167